I made a mellow S2+ triple shorty with gen.2 92CRI 3000K Oslon Square leds a while ago with a 2x7135 driver (700mA) as a reading light for my girlfriend, and it occurred to me today that with its very flat output it is a good light to measure what the actual light loss is from this common BLF-mod. The typical Carclo 10507 TIR triple optic claims 10% loss, but that is measured with an XP-E size die, and there's the rim at the front that blocks some light too. And I like to use the glass lens to protect the optic from scratching.
I found a piece of flashlight head in my junk pile that had the wrong threading but could conveniently be forced between pill and battery tube so that a bare triple head was constructed for maeasuring the primary amount of light in my integrating sphere. Both set-ups have exact the same regulated current and I suppose the leds put out the same amount of light.
So here are the numbers:
Bare leds: 241 lumen
Assembled S2+ with clear Carclo 10507: 188 lumen. So there's a 22% loss
Assembled S2+ with frosted Carclo 10511: 175 lumen. 7% loss going from clear to frosted
Assembled S2+ with frosted Carclo 10511, stock glass lens added (no AR coating): 164 lumen. 6% loss from the lens
So going from bare leds to an assembled S2+ with clear triple optic and non-ARcoated glass lens will cost you 27% output. (edit: Nichia 219 leds and probably XP-G2 as well do a bit better in this optic, see post#20)
I do not think if measuring the same for a black S2+ host will obtain any different numbers. Using other leds will give slightly different numbers I guess, because the light will enter the optic somewhat different but I doubt the difference will be much.