There are many vendors of pre-made solutions, each with a slighty different features/tradeoffs/etc. Here are some starting points:
Most LED drivers are current-control (variable voltage) rather than voltage-control (variable current). Current-control is more complex and often less efficient, but it helps drive the LEDs consistently across a wider range of situations. With voltage-control, when the LED gets hot and resistance drops, the current goes way up and the LED can fry. You can dodge by under-driving the LED (e.g., half the rating) so even if the current goes way up the LED is still within spec. Current control lets you drive the LEDs at higher current without thermal runaway.
As a rule of thumb, the closer the battery and LED voltages, the higher the efficiency of a given driver. However, different driver designs have widely-varying efficiency, from under 50% to over 95% efficency. Thus, driver A may give better efficiency with a big voltage difference than driver B with a small difference.
Beware that efficiency ratings are usually peak, and different driver designs have widely varying efficiency as you move away from their ``sweet spot''. So, for example, an ``88%'' efficient driver might give better efficiency in your specific application than a ``92%'' efficient driver that is being used outside of its ``sweet spot''.
As the batteries fade, you eventually run out of ability to drive the LEDs at full brightness. Some drivers will cut out. Other drivers will keep driving -- the LEDs get dimmer and dimmer, but you can still see. And, you can still damage the batteries.
Most drivers are smaller (and lighter) than they appear in the pictures. The TaskLED ``Fatman'' regulator, for example, is about the size of two U.S. nickles -- but weighs less.
LEDs dim as they age, and dim faster when they are hot. Keeping LEDs cool improves both their service life and their efficiency. Under-driving them (e.g., run a 1W LED at 0.5W) brings modest efficiency gains. Over-driving them (e.g., 1W LED at 1.5W) gives little extra light compared to the extra power you burn.
The human eye response is logarithmic -- so it takes pretty big steps in power to make a noticible difference in brightness. The flip side is you can cut the power a lot and still have useful ight. I have an experimental light running at 8W/4W/2W (4x Luxeon-III driven from a 9.6v nominal NiMH battery pack and a TaskLED Fatman regulator) and the 2W setting useful in lots of situations even though it draws relatively little power.
As an aside, there are various supplies of good optics that take the wide-dispersion pattern of an LED and make a nice spot.