Thanks bykfixer for the run time test. It's clear that unlike many lights, including many Maglites, this one is going for flat, regulated output until the cells give out. I suspect the 3C does the same. Many lights as I'm sure most are aware slowly or even rapidly step down the output. This in a sense plays games with the rating system because you can claim a very high brightness, dim to say 11% after 1 minute then run for hours at this refused output level. The XL50 is VERY guilty of this sin. Hence 3AAA batteries seem to produce almost as much output and runtime as the 3C model.
http://www.led-resource.com/2010/10/maglite-xl50-led-flashlight-review/
As for the much longer 3C run time, I think that is a case of just a very long tail on output vs time plot. Note one of the comments at the end of this review
http://www.led-resource.com/2015/04/maglite-ml300lx-review/
A poster noted that the 2D light went 33 hours at the low (143 lumen) setting while the 3D made it to 143(!) at the same output level. Well if you look at the plots you see that the 2D stayed flat for 11 hours vs 33 for the 3D. The 2D clearly has a less efficient boost converter (understandable). But that doesn't explain 33 vs 143 hours to get from 143 to 14.3 (10%) lumens. The review addressed this in a comment after the article. Basically they said that with 3 cells you don't need a voltage boost to drive the LED at a low level. So the system can run for a very long time with low cell voltage. However, when you consider the 2D system you need enough voltage to operate the boost converter. So you effectively hover just above 10% for a long time with the 3 cell but under 10% for a long time with a 2 cell. If the ML25 is like the ML300s then the 3C likely has 2x the flat run time and a very long, low output tail.
Incidentally, while this seems like kind of a cool light, my local Walmart has the ML50L 3C lights for $33. Not much extra for a light with huge output (that drops fairly quickly according to the ML300 plots) but also has various modes.