Today I took some time to do some further testing on the XP-L.
I had bought two of the same leds (XP-L V6 2C, from Illumn.com), in one piece of cut reel, so they were right next to each other. The first one was destroyed in the crash test of the OP, and today I used the second one to repeat the test in order to find out to what extent the results are reproducable. Of course if the results from the two leds differ much I can not tell what part comes from variations in the method (i.e. it was 5degC warmer today than it was last time) and what part is variation in the led.
I used the exact same equipment, even the same star and led-wires from the first test, I just removed the faulty led and reflowed the other one. I tested the integrating sphere with one of my reference lights to be sure it read within 1% the same as it did last time (it did).
Here's the result, I plotted the results of the second led in the same graph as the first one, for comparison (I left out the XM-L2 3C results from the OP, just the two same spec'ed XP-L leds). I measured up to 8.5A this time to save the led for further use:
The second led measures slightly different from the first one: the voltage reads about 0.05V lower at most currents, with a maximum difference of about 0.1V, the output at most currents matches almost exactly the first led, only above 6A led#2 starts reading a bit lower than led#1. So this is not an efficiency difference (that would affect the entire curve, not just above 6A), but it looks like led#2 shows slightly worse heatsinking than led#1. Perhaps it is the led, perhaps it is the reflow that was not as perfect, I don't know: the difference is not so great that I feel an urge to find out.
What I am happy about is that this integrating sphere system seems to perform consistantly, I am really beginning to trust the thing :-)
While I was at it I thought it would be nice to compare this led domed versus dedomed. So after the test, before it was cooled down, I took the copper mount out of the sphere and with the led still hot I dedomed it by lifting the dome off from the side with a scalpel. It wasn't the cleanest dedome I have ever done (usually I do dedomes under a stereo microscope, I had to do with a loupe this time), but the die was clear from silicone afterwards, so the dedome was effective (the led is illuminated from the side to show remaining silicone well):
(stereo pic)
The dedomed led went straight back into the sphere for direct comparison with the domed state. I was very surprised to find this:
(sorry I'm not very good in making chart dimension the same everytime, they differ a bit from the first chart)
So what has happened? The concensus is that dedoming lowers the Vf (that is at least a suspicion) and costs at least 20% output (that is measured by many). This led however dedomes without light loss (that is: nothing under 5A, only slightly above that) and the Vf stays dead-on. Tell me what's going on here?, it looks like the performance is the same, just at very high current the heatsinking seems to have suffered somewhat, but I can explain that if I must by the better heat dissipation properties (surface area) of a dome vs direct air.
The led was really dedomed well (see pictures above), and I checked the light colour compared to a few other leds: xml2 3C, Nichia 219B, an xml 'cool white' led from the spare leds box (you can find out which is which :-) , the XP-L is on the right side in both pictures):
Pictures do not show tint well, but by eye the dedomed XP-L looks like a little under 5000K and more yellow than both the 3C and Nichia. So it definitely looks like the tint shift that you expect from a succesful dedome.
I can think of three theories, all quite unlikely:
1) the tint shift to a warmer tint leads to less output but at the same time the luxmeter is more sensitive to the warmer light, and those two effects level out. This is highly unlikely, the wacky wavelength response of this type of luxmeter can not account for a huge 20% measurement error for a mere 1000K tint-shift. And besides, the by many observed output-loss upon dedoming was very probably done by the same type (or the exact same) of luxmeter as is in my integrating sphere.
2) the slightly altered 'beam profile' of the dedomed led as compared to the domed state causes the 20% measurement error. This is not true, shining a focussed aspheric flashlight around in the sphere at all possible angles causes very little variation in the reading, maximum variation is 2%, only projecting the led image straight onto the baffle shows a reading 10% less.
3) this is the conspiracy theory : the XP-L's that Cree sells at the moment have domes that do not attach well to the die: domed leds behave like dedomed leds in light colour and output. It would explain why I do not find an efficiency for a V6 that is any better than a T6, it would also explain why Old Lumens notices such a warm tint for his XP-L in that HD2010. I failed to check the tints of my XP-L's beforehand, so I do not know about them. How's that for a theory :-D
There must be theory 4 that explains it all, like my measuring methods are flawed after all and such. Enlighten me :-)