I just looked it up...Note 4 page 6,
http://www.lumileds.com/pdfs/DS51.pdf
"The forward voltage of the Luxeon K2 LED will reduce by up to 0.30V at 350mA during the first few hours of operation
after SMT reflow."
Emphasis, mine. Notice that this is after reflow is done.
But even more important, keep reading:
"Due to this effect, Philips Lumileds
recommends current source drive for consistent and reliable performance. Cross connected series/parallel arrays or voltage drivers which could result in current hogging or variation in drive
current are not recommended."
Emphasis mine again.
For a T bin K2, see page 7:
"The forward voltage of the Luxeon K2 LED will reduce by up to 0.50V at 1000mA during the
first few hours of operation after SMT reflow. Due to this effect, Philips Lumileds
recommends current source drive for consistent and reliable performance. Cross connected series/parallel arrays or voltage drivers which could result in current hogging or variation in drive current are not recommended."
I have seen this same Vf drop effect, after SMT reflow, and without SMT reflow, but I wanted to characterize things closer before I talked about it much... The Luxeon I and Luxeon III also show a Vf drop with time. Not all brands of high power LEDs show this Vf drop with time.
There is an older thread where I talked about this with one of the scientific researchers at LumiLEDs (Michael Krames- Manager Advanced Laboratories, Lumileds Lighting), and it was due to several mechanisms.
We also talked a bit about the subtantial increase in LED efficiency when current dimmed, and the loss of efficiency when PWM dimmed (for white, blue, green, cyan, dental blue, royal blue), and why it doesn't occur in their reds.
It is just much worse in the K2, than the Lux I and Lux III. But when you start driving a Lux III up at 1,000mA, they come close to changes one would find in these K2s.
There is an old thread from 2004, where we were talking about this issue, and you can find some of the charts here:
http://candlepowerforums.com/vb/showthread.php?t=60217&page=4&pp=40
evan9162 has some better charts in some other threads, dealing with Vf shift in Luxeon I, Luxeon III, and Luxeon V devices, but I'm not sure where those are hiding right now....