You are quite right, the circuit has to be tuned to the Vf of the LED in use. But I never make more than one of these at a time anyhow. I've read specs for some drivers (DX SKU25517) that state different current for different LEDs - maybe they use this circuit!
As you noted the open circuit voltage is ~5V. This is a useful side effect that acts as an over-voltage limit for step-up converters.
So, thinking about it, a possible solution would be, instead of the 6.04K resistor, a cheap standard 1n400* diode. With the 26.1k resistor acting as a pullup, the fraction of ma through the diode should give it a hair under 0.7v drop, raising the .3v across the sense resistor to right around where the controller wants its feedback voltage. Slight fine-tuning could be done by playing with a resistor in series with the diode (adds some voltage dependency, but much less than with no diode), or with the value of the sense resistor. (keeps it constant-current, but might be more work). With a .1 ohm sense resistor, it would be ~2.3A. With a .1 in parallel with a .33 as the sense, you'd get a nice 2.9A...
I finaly made proof of concept and it's OK. Unfortunately there was one casulty - because I made wrong connection. I've put one of leds to direct drive 7V+. No flush or whatsoever. It just passed away.
Nevertheless the whole circuit works with same current across two led groups (1 led = 1st group, led 2 and 3 = 2nd group). Modes are also working across all leds. I haven't had time to make photos, jet.
Well, it was only 1.4A this time. Next time with 2.8A, I'm going to be more careful and I'll make shure everything goes well, no solderless connections.
It was interesting anyway as I got 7+ V to the driver for a short time of few seconds and it survived well. Don't know what the current was at that moment, but two paralell leds were lit up. Yeah, I'm in the club
A question for you. I've measured different currents at leds in paralell. The one closer to the driver took 0.8A while second one took 0.5A. Can it be because of shorter wireing of the first one? Should I take care wires to have the same length for both leds? Or alternetively I should double wires to second one?
Paralleling LEDs is a bad idea, for exactly that reason. LEDs are designed to be driven at a constant current, and the forward voltage may vary. Running two at the same voltage will give different currents and brightnesses.
I undrestand, but you didn't answer me on wireing to improve situation as much as I can. In real situation I'm going to have 3 XM-L from (hopefuly) same lot already munted on PCB. I can expect very similar performace and I can choose best match of two from three.
First off I would double check all your connections. You may want to solder the wires rather than alligator clip them. If the difference in current sharing remains and isn't acceptable then you have a few options.
Put the higher current (lower Vf) farther from the driver (longer wiring = more voltage drop) Don't be disappointed when the doesn't work... the length of wiring shouldn't be an issue.
Add a wee bit of resistance in series with the higher current (lower Vf) LED. By wee bit I mean somewhere in the range of 10 to 50 milli-ohms. This likely can only be optimized at one current, resistors are linear while LEDs (like all diodes) are non-linear devices. This approach will lower system efficiency a bit.
Buy or select LEDs with matched Vf's. This begs the question of "how matched?". If you will be buying another LED to replace to one you poofed, maybe it will be a good match with one you already have. (It could happen...)
Live with it. (Embrace the imperfectness of this imperfect world. - Zen approach) The difference may be hard to tell, and you've already accepted that the series LED will carry the sum of the current of the other two LEDs.
Bushytails is correct - parallel LEDs causes problems with current sharing. The way the four LEDs in an MCE may be paralleled is because Cree matches the Vf's through process control and testing.