It might get a little more complicated than that, that there’d be a similar lag when turning off, but the main issue is in fact getting the bugger to turn on in the first place when the PWM period for 1 count is about as long as you want to strobe the 7135s.
It’s like turning on a bank of tube fluorescent lights that flicker a bit before catching. That lag is what causes the issue when you just want to turn the lights on for, say, 1sec, when it takes about 1sec for them to catch.
And the bigger problem is that there’s variation elsewhere (7135s, output drive for the Tiny13s, etc.), that some combos will work fine and give you nice low firefly mode, whereas other times it won’t turn on at all, or start fine in firefly but not at all when stepping down from turbo to firefly, etc.
I followed the thread for a while, on setting PWM values on “biscotti”, and it got really complicated.
For a Li-cell, the best firefly mode is probably just a dedicated line through a resistor, not PWMing it at all. Besides, as I can attest from my Quark, lowest modes tend to be the greenest (yecch). That’s why PWMing is actually preferred over linear. Linear brings its own issues such as nasty color-shifts in any LED (even monochromatic, not just white). Even 2A vs 2mA is a 1000:1 range.