I've been trying to understand it as well. I've been analyzing the layout, and PD's reply to my question above. The default layout of this board is like a FET+1 driver -- drives only 1 emitter, but instead of 2 paths the MCU controls for the LED output (a 7135 and a FET), there are 3 paths: one single 7135, a bank of 6 7135's, and 1 FET.
The limitation we have to work around with the ATMEL tiny MCU's is that we can do PWM's on only 2 output pins, but with PD68's design, the FET is simply switched ON or OFF with no PWM's because you don't need them. You can use the bank of 7135's for all the in-between modes, and use the single 7135 for the lowest modes.
My preference for a mode set is 5 modes w/moon, so in the Narsil firmware, you choose a 4 mode set plus you turn on moon, so modes are as follows:
Mode 1 (moon): 3 or 5 PWM value on the 7135
Mode 2 (~2%): 30 PWM value on the 7135
Mode 3 (~10%): full output on the 7135
Mode 4 (~40%): 80 PWM on the FET
Mode 5 (100%): full output on the FET
So you can see from above, the first 3 modes are actually on one 7135, and the top two modes are only on the FET. TK's spreads it out, sharing the 7135 ad FET through the mid range modes.
So on this triple down driver, if I wanted to implement my 5 modes on it, I'd use the bank of 7135's only for mode 4. Six 7135's is 2.1A, so that would be 42% for a 5A max FET. However, as you know max amps on a FET varies based on the LED, cell, cell's charge level, etc., so these percentages will vary. The 7135's will give you something more solid, but still at some point, 2.1A will not be achievable depending on Vf of the LED and power/charge of the cell.
The advantages of this implementation is we are getting very efficient on all modes because PWM's on a fully open FET is pretty inefficient - it's more wear/tear on the LED by cycling 5A at a high rate, you get the tint bias for the 5A, and lumens/amp ratio is down. I can even see using only 5 7135's instead of 6 to get closer to a 35% level for your next level below hi/max. Another thing is you can treat using a full output on the 7135 bank as your hi mode, and implementing the full FET as a turbo. We have a lot of flexibility here with the 3 channels of output and can achieve very close results to a true non-PWM driver, like the LD-1/LD-2.