In their method, the light that would benefit from collimation to the target is instead already utilized to boost the source intensity. The trade off will not be beneficial over direct collimation toward the target. It's like running on a treadmil with a generator to produce energy to get somewhere, when you could have just run there to begin with. We can kick around that light all we want, but it's not going to be more efficient than directing it straight toward the target from the get go. I wish I saw it differently because it would be a fun new avenue for sure.
I would disagree with that. First of all, note that with any collimation method, reflector, TIR or lens, the light won't go "straight" at the target as you say, it'll go slightly off to one side, more so for a reflector than with a lens, hence why when someone wants extreme throw with an LED at a certain size, they go for an aspheric and not a reflector. Have a look through the datasheets that various manufacturers put out, then compare with the figures for aspherics that people here have gained from experimental data. Perfect collimation of all light gathered (whether from a reflector or an aspheric) can only come from a point source. Failing that, Aspheric > TIR > reflector, even when the latter two are optimised for throw.
So let's take an LED with an aspheric. The aspheric captures all light emitted up to, say, +/- 30 deg off axis. Assume we have 300 lumens coming out of the LED, 150 hit the aspheric, while the other 150 miss it and are completely wasted. Assume transmission losses to be 10%, that gives us 135 lumens OTF.
With this collar, it captures some of the
wasted 150 lumens and redirects it back at the LED &c. as described earlier, giving the 58% increase in surface brightness. Apart from the obvious decrease in total output due to inefficiencies in the collar, the main tradeoff is that the viewing angle has now been greatly reduced, eyeballing the image on their site, I'd call it from 180 deg down to 60 deg.
If we were using a reflector based system, this would throw a lot worse, but see my previous note about throw and reflectors vs aspherics. Our example aspheric, however, doesn't use any of the light past +/- 30 deg, so we have a gain in overall throw due to the greater surface brightness. Note that if the aspheric only captured light between, say, +/- 20 deg, then there will still be a large amount of light lost, hence Saab's comment about matching the lens.
So with the LED, aspheric and everything else being the same, therefore we get 150 * 1.58 = 237 lumens hitting the aspheric, while only 63 lumens is lost. Taking transmission losses into account, that's 213 lumens OTF.
Short version: Viewing angle down, surface brightness up.
Will OTF lumens be lower when compared to a reflector or TIR (especially the TIR)? Gut feeling is that it will be, but I can't say for sure. But will throw increase? I believe yes for a tuned setup, for the reasoning given above.
Enginyr and thezug, my opinion is probably worth as much as a black reflector, but I'd like to second what bshanahan14rulz said, if you come up with something like this for Cree's XP series, I'd buy one without hesitation.