So you want me to carry a battery pack AND a supply of ice? No thanks. LOL 
3X18350 in the short config would work, but I have no use for a light that runs 15 minutes @ 2000 lumens and no other mode.
This has no use, it's for fun...hobby. Actually at ~ 3A it is so darn bright that it can give my 65W Ebay type HID in low mode a run for its money (just that the central hotspot of the HID is a bit brighter still).
Do you think all of us really use the 20-160 lights that we have? LOL! Just a couple of rare occasions. The most used light in my collection is the Philips brand 3 LED passive IR dectector light in the toilet, its for me to aim my pee properly. Costed me 19 bucks, well i got it long ago before discovering DX/aliexpress LOL! Supposed to work with 3 x AAA but I modded it to do 1 x 10440 (coz I ordered 2 x 10440 in DX but they sent me 2 x 2 = 4 LOL!). Some stick it in the cupboard too.
This is good too.... http://www.dealextreme.com/p/8-led-2-mode-infrared-sensor-motion-activated-white-lamp-2-aa-35465
Is there one particular version that is going bad? I've lost track.
The reports of my skyways demise appear to have been greatly exaggerated. My boy sent me a text last night saying it burned up. When I got home in the morning I found that it was the flashlights low voltage cutoff. He tested the batteries last night and thought they were ok because they were at 3.06V. He’s just a kid and he at least thought to check the batteries so that’s good. He’s happy now and so am I knowing that the low voltage cutoff is working.
What is the low voltage cutoff for this flashlight? and How many batteries are you using?
If you are using 2x18650 and if the flashlight cutoff is 3 volts then I guess the disconnect was at one of the batteries protection circuit, unless your batteries are being drained to ~1.5Volts each one.
Good to know that your light is still ok Langcjl.
I think generally if you do not gun the light for more than 10 mins from resting it should be really fine. Initially I think I said it could be gunned indefinitely, better not do that, unless you are really sure what you are doing and can ensure that the light is cooled in ambient air nicely, ie your temperature is really cool like below 10 deg C. If you have a breeze and it is cool then it is pretty ok. Most of you here I see experience freezing conditions during winter!
The issue is some of us live in Florida and tropical countries. The hottest night we get here can hit 31 deg C.....in some other places like Bangkok it could even be higher, nearly all year round.
Both batteries were at 3.06V so it was the flashlight that shut down. I think the batteries will drain a lot more before the protection trips.
I did another crazy experiment and had stuff recorded. Late last night one of my family members came home with some ice-cream cones which were treats from the company and in the box was some dry ice which has not sublimed yet. Well actually I used to buy dry ice for the fun of it, all 30kg of it delivered in styrofoam box, but it is pretty expensive I think about $80. (it is difficult to get liquid nitrogen here, i think it is not available for "public use"). Like i said I had one thick cable for the + and I soldered the -ve to the body of the flashlight and then i use several smaller dia cables...what 24-26awg, to connect to my battery.
This time round I cooled the thing with ice cubes first and so it was nice and cool. Packed the insides of the head (ie that cavity that houses the driver and back of the emitters) with dry ice powder. I had to keep the LEDs running at "low" juice with 2 cables (about 3A) to keep it warm because the front lens was fogging up. Nah...no go, i had to take the lens out, it was still too cold and way below the dew point! LOL!
The reflectors are pretty ok and didn't fog up as I managed to keep a 20" industrial-type circulator fan blowing at it. I think in the USA you'd come across the "Vortex" brand, but it is a Taiwan knock-off called "iFan". LOL! (but seriously it works the same, i measured the wind speed airflow with an anemometer at the shop.... LOL!)
Everything stabilised, i proceeded to pack the exterior up with powered dry ice as well. But I think the internal cavity is doing most of the work coz that is closest to the stars. Note : I guess 6 nines copper and mounting the LED directly on it would be much better, but i don't have that luxury. Besides, dry-ice cooled aluminum and stars still takes a while to warm up, plus with such a large temperature gradient between the dry-ice and LEDs, the heat just gets pulled away like there is no tomorrow. BRUTE FORCE - I LIKE.
What I did is that i have a PSU and i'd believe the current is 5.2A (not sure why was it fluctuating somewhat...probably heating up slowly at the junctions and failing), and coupled with the crazy cooling it managed to survive just long enough for me to get a lux reading.
485 lux @ 8.17m. 32373 lux @ 1m.
Based on bro's engerolitis IS reading of 1600L OTF for stock (which corresponds to 260 throw lux at my 8.17m range), this is 485/260 lux = 1.86 x 1600L = 2984L OTF.
Another way, I have my UF U80 XM-L which does 65 lux stock in reflected bounce stock measures 2.55A at the tail. That's 800 lumens emitter. Stock SR3800 = 169 lux reflected = 2080L emitter. 1.86 x 2080L = 3868L emitter from this overdrive.
But seriously I do not know if i damaged the LEDs, so that's not knowing if the efficiency dropped.
PS. I am using that DX meter, which is known to under-read by about 10-20%.
I think it's about due time someone did peltier cooling for flashlights. Perfect for this application since you can share the PS (~10-12v).
It is interesting to note that with 5A to the lights, without active cooling, the heat of the LEDs actually cause output to go down. Play with it in the LED calc spreadsheet.
But with active cooling the reverse happens, ie a solid cold source which pulls the heat away, you can cause the junction temp to drop and that is the only way in which the emitter lumens go up.
ie at 5A, and at -42 deg, 1285L emitter is doable. 1285 * 3 = 3855L emitter. Pretty in-line with my readings.
Vaporising/Subliming dry ice has a specific heat capacity which is twice than that of ice. Also dry ice has 1.56x mass than that of water/ice per unit volume. And it does that at -ve 78.5 deg C, while ice is at 0 deg C. Hence for that same cavity, it has packs a total absolute punch of over 3X the cooling potential of ice/water. So with that amount of pelleted dry ice in the cavity and also outside, you need a lot of heat to use all that up.
When I was still playing with mod cars, we were packing dry ice on the intercoolers for those short sprints so as to manage the heat soak prior to that. (remember over here it's damn hot)
My SR3800 is alive again! This time direct drive from 3 x 18650. I ripped out the driver PCB and used the backing only, just 3 mins worth of soldering and off she went.
I purposely discharged some 18650s to about 4.00V and used some protected cells which are not "that powerful", ie don't really supply good current on my UF-980L. Namely the PALight 3000mAh ones that I got from Manafont and i mixed with another cell. Seriously these are all short bursts, no issue.
I mixed and match my Trustfire 3TF so that I can get 3 x 18650 length.
So what I get is 342 lux of intensity at 8.17m, or 22828 lux @ 1m converted. Add 10% (my meter under reads) and it is about 25k. About 2.4A on the meter, not sure how much voltage drop so i will never be sure of the exact figure. Sky Ray SR3800 stock is 260.
The max I have gotten with dry ice cooling was 485 lux @ 8.17m powered by a 7AH SLA (plenty of juice, little sag) still connected to the charger, i got over 5A on the meter. Or just over 35k 
(already added 10% for compensation)
This is 342 lux, vs 210 lux for the Trustfire 3T6. (6250k manual WB) Relatively speaking
I put kai driver in my skyray 850/4500, it does 4A at tailcap and really nice light level :) Only you have to put driver in a thermal grease, it gets hooot, but flashlight made 5min test without problems.