2000 "Winners" roller clutch casstte hub. Had a reputation for slipping under load, and for locking up when trying to coast.
Roller clutch is HFL2530; at least one maker rates it for 66 N-m [XB09]. Note lack of seals on roller clutch. Hub weight approx 330g.
For comparison, consider a 100kgf rider using a 175mm crank, standing with their full weight on the pedal using a 1:1 gear ratio such as 34:34 or 22:22. That is 100 kg * 9.8 m/s2 * 0.175 m = 170 N-m or about 2.5x the nominal load rating of the roller clutch. Worse, roller clutches are typically rated when pressed in a hole in thick steel. The Winners thin aluminum will be substantially more elastic and might be dented by roller loads; both would cause further derating.
The lack of seals makes the roller clutch subject to grit intrusion, which may jam the rollers and prevent coasting, and grit may accellerate denting that can prevent both engagement and coasting.
A short history of roller clutch freewheels: they fail.
A slightly longer history:
A roller clutch is a set of rollers arranged around the outside of a smooth drum, and inside of a ring of ramps, one ramp per roller. When coasting, the rollers roll "down" the ramps, gaps open up, and the drum to turn freely. When pedaling, the rollers roll "up" the ramps, close the gaps, and the rollers jam — thus transmitting torque. With small clearances, "jamming" engagement is nearly instant. And since the rollers roll on a smooth drum, coasting is silent.
For bicycles, the appeal of roller clutches includes (a) they are quiet; and (b) they they have low lash, which is to say, the freewheel engages as soon as you start to pedal — which feels nice, and is useful for riding trials. Some claim low lash is useful for other cycling as well.
And, roller clutches are a standard industrial part, used widely — including they are used in most car automatic transmissions — so it seems like they should be easy to use for bicycles.
The basic problem with roller clutches is they generate a lot of hoop stress. When the rollers jam, each is like a wedge, trying to split apart the ramps and the drum. The wedge is like being on the wrong end of a lever, so the clutch has to be much stronger than the pedaling load it carries. Data sheets for industrial roller clutches of sizes used in bicycle hubs show torque specs much lower than ordinary pedaling, and industrial rating is for clutches pressed in a thick and heavy steel ring, whereas bicycle hubs use a light aluminum ring.
Some users report failures where the rollers slip when pedaling hard. This may be due to flattening of the ramps or dents in the ramps, allowing the rollers to reach the end of the ramp without jamming securely. Others report a tendency to jam, probably due to denting which keeps the rollers from moving freely in to the coasting position. Finally, a few riders report the clutch jams in cold weather, though the hub appears to be undamaged and the jamming goes away when the hub is warmer.
Some historical examples:
The earliest bicycle use of roller clutches may be the 1903 Sachs Torpedo coaster-brake hub, see http://yarchive.net/bike/torpedo_hubs.html as of 2009/05. According to Hans-Joachim Zierke, those had a service life of 100,000km and oversize rollers were available to rebuild them to extend the service life to hundreds of thousands of km. They were also relatively heavy hubs, even subtracting the coaster brake, and they were typically used with a large chainring and small sprocket, so got somewhat less torque than is typical for hub for derailleur gears.
There are stories some company tried roller clutch freewheels in the late 1970's or 1980's, and after little use they started slipping under load.
In the mid/late 1980's, Bicycleville/Bushido Racing said they would make a roller clutch freehub. They had a demo bike with 30cm front suspension, 25cm rear suspension with on-the-fly adjustment; disk brakes; pneumatic seatpost for on-the-fly height adjustment; and a claimed weight under 30 pounds. This was when Rock Shox was selling a fork with under 5cm of travel to put on bikes not designed for suspension, because no mainstream bikes were designed for suspension. Bushido's demo bike probably changed the bike world, although the only thing reportred sold were conventional frames with rectangular-tubing chainstays.
Shimano's 1990's LX RO-80 "silent clutch" hub was perhaps the most successful roller clutch cassette hub to date. It was heavy by today's standards (over 500g) but the roller clutch withstood significant casual use. Trials riders liked the "instant" engagement enough to make it worth replacing the roller clutch portion every few months. Unfortunatley, Shimano stopped selling replacement roller clutches, and the "silent clutch" uses a different spline than other cassete bodies, so a damaged silent clutch cannot be replaced with a conventional ratchet.
In 1994, Cane Creek made a few roller clutch cassette hubs using carbon fiber instead of heavy steel. According to http://www.firstflightbikes.com/1994_Salsa_24.htm as of 2009/05/01, "The hubs never really made it into production due to the relatively limited torque capacity of the carbon band to hold the roller mechanism in place."
In 2000, "Winners" offered a cassette built around a roller clutch. They were light for the day and felt good, but reportedly sometimes slipped under load and locked when coasting. The hub lacked sealing, and grit intrusion would tend to make increased point loads, possibly exagerating both problems.
Also in about 2000, True Precision Machining (http://www.trueprecisioncomponents.com as of 2014/10) introduced their "Stealth" hub using a roller clutch. They reportedly place the roller clutch in the body of the hub shell, allowing larger roller clutch. From http://www.cyclingnews.com/mtb/2009/apr09/seaotter09/?id=/photos/2009/tech/shows/sea_otter09/sea_otter099/Stealth_clutch as of 2009/05, it appears to use a custom roller clutch, rather than an industrial clutch. Compared to industrial units, the custom clutch appears to use fewer rollers of larger diameter. A large roller probably helps prevent local indentation, which seems to be one failing of industrial clutches pressed in an aluminum carrier. Despite the custom clutch, one reviwer at http://www.mtbr.com/mfr/true-precision/hub/stealth/PRD_358018_127crx.aspx as of 2009/05, reports lockup of a hub under three months old, and lockup of its replacement. Also, as of 2014/10 it is listed at 496 g and US$415 for an MTB hub with IS disk mount, compared to an American Classic "Disc 225" hub at 225 g and US$276 (MSRP as of 2012/09, but 2014/10 retail prices seem comparable).
Shimano's Alfine SG-S500 internal-gear hubs also use roller clutches; like the Torpedo, the hub is substantially larger and heavier than conventional cassette hubs, and is typically not used by vigorous riders.
In 2014, Efneo (www.efneo.com as of 2014/10) announced a roller clutch hub claiming (as of 2014/10):
In traditional [roller clutch hubs] the internal hub rollers where worn out by continuous friction against the external ring — this made the rollers smaller and smaller, what caused the hub to slip over. [...] The weakness comes from the construc-tion in which the rollers are grinding all the time that the cyclist is not pedaling. After a relatively short time they are worn-out and do not engage effectively when a cyclist pushes hard [...].
Efneo freewheel works differently when not pedaling. The rollers are not grinding at all so the rollers do not wear out nearly as quickly as other roller freewheels did.
There is no evidence from other designs that "coasting wear" is a significant problem — both that designs such as the Sachs were successful, despite being conventional; and at the same time, "coasting wear" must be slow as the loads are small, yet designs such as the Winners failed quickly, before any significant wear could occur.
In 2015, a company operating as "Caron" advertised a "silent twin clutch hi-tec cassette hub" ([HERE] as of 2015/08), claiming
The Silent Twin Clutch Hi-Tech Cassete Hub is designed based on our new technology 'clutch bearing'. As the back-lash is almost 0 (0°-0.2°), no-load rotations are prevented and 100% delivery efficiency is achieved in chain operation. As this hub does not produce any noise (pawl noise), it satisfied riders in terms of speed and riding quality.
Also, this hub has been developed with 20% higher intensity than that of traditional bicycles and can thus be applied in all types of bicycles. [...] The clutch roller of the Silent Twin Clutch Hi-Tech Cassete Hub requires special steel and high hardness (HRC55), and must be processed at 0.01 tolerance using a centerless grinder.
What that all means is unclear, but the text suggests a roller clutch hub, and as "you can't cheat physics with wishful thinking", it thus seems likely this is another failure-prone hub.
The main goals of roller clutch hubs are low noise and quick engagement.
Many pawl-based cassette hubs are loud, but many freewheels are so quiet they are inaudible over road noise when coasting. One difference is most freewheels use light pawls, while many cassette hubs use heavy pawls. Although some freewheels have a reputation for failing in heavy tandem (Sachs, is one example), others were both quiet and apparently durable in tandem use (Shimano and SunTour, as examples, though SunTour freewheels are often damaged during removal after heavy use). Older Shimano cassette hubs use heavy pawls and are quiet, so pawl weight alone is apparently not the deciding factor. These examples suggest pawl-based cassette hubs are loud is due to design choices, but they do not need to be loud.
Many pawl-based cassette hubs rotate 15° or more between ratchet "clicks". Chris King "ring drive" hubs have 5° clicks and are popular for trials; they are, however, loud, and it is not obvious they can be made quiet. Hope Pro 2 hubs use two offset pairs of pawls in a 24 tooth ratchet for 48 clicks or 7.5°. Again, they are loud, but, as above, careful design might make them quiet yet give good service, at least outside of trials. Industry Nine sells (as of 2009/05) a cassette hub with large-diameter ratchet, where each pawl has three steps, the ratchet has 60 teeth, and six pawls are arranged in two offset pairs [CN09], yielding 3° steps. Durability and noise level are unknown, and it might not be possible to make stepped pawls quiet. However, the hub suggests a hub like the Hope Pro 2 could use three pairs of pawls to achieve 5° steps.
Freewheels typically used two offset pawls, so only one at a time was engaged. There was, however, sufficient space to use four pawls as two pairs, possibly even six pawls as triples or three pairs. Jobst Brandt argues [Br03, Br04] that at any instant one pawl is carrying most of the load, so there is little advantage to having several pawls engaged simultaneously. The observation is bearings have play and parts are both elastic and not perfectly concentric, so some configurations rest the load on one pawl and the bearing. However, free-body diagrams suggest having two or three pawls engaged simultaneously improves load carrying capacity, and even if the load is sometimes carried on one pawl, experience suggests having pairs or triples of pawls engage together improves durability, as load is shared between several pawls, even if it is shared somewhat unevenly. Thus, it appears some durability problems of (quiet) freewheels could have been addressed without using heavier (noisy) pawls.
A sprag clutch (also called "Sprague" or "Formsprag™") is similar to a roller clutch, but instead of using rollers, ramps and one drum, it uses two concentric drums and "rollers" that are shaped similar to a figure "8" or capital "O". When coasting, the rollers "fall over" and let the parts slide by; under load, the rollers "stand up" and jam. Sprags do not need as much space to move and so can be packed more closely than rollers. They are also typically larger than rollers and so have more load bearing surface and less tendency to dent. For these reasons, sprag clutches tend to have higher torque rating than roller clutches of similar size. Sprag clutches are also radially thicker, making them hard to fit inside the sprockets, as on the Winners hub; but they might fit inside the hub shell. They may also have higher coasting drag. Industrial sprag clutches are not obviously available in bicycle sizes, and as best I know have not yet been used in bicycle hubs.
A "Micrometer springless clutch" uses rocking crescent-shaped pawls of approximately half-moon shape. Instead of springs, geometry causes them to rock when coasting, and to engage when pedaling. They are essentially silent, but prone to failures both from wear and high load [Ha02].
[Br03] Jobst Brandt, "Re: KING HUBS -- LOUDEST ON THE MARKET?????",
[Br04] Jobst Brandt, "Re: American Classic hub design problem". From http://yarchive.net/bike/freewheels.html as of 2009/05.
[CN09] Cycling News, "Sea Otter Classic Expo Monterey, California, USA, April 16-19, 2009", http://www.cyclingnews.com/mtb/2009/apr09/seaotter09/?id=/photos/2009/tech/shows/sea_otter09/sea_otter099/Industry_Nine_drivers as of 2009/05.
[Ha02] Brian Hayes, "Sturmey-Archer SW Three Speed Bicycle Hubs", http://my.ohio.voyager.net/~bdhayes/sa/sw.htm as of 2009/05.
[XB09] SUP Bearing, http://www.zxz-bearings.com/needleclutch1.htm as of 2009/05.