Hmm... so good and bad. At 3.0V input, the new setup (v chip, Ta cap, low leakage diode, tuned voltage divider) is able to measure off-clicks over 3s in length, (I'm pretty I measured it closer to 4 once, but haven't tried to reproduce that) using only 47uF of capacitance, and does it even with high overall bleed resistance (4.3kOhm). At that voltage and bleed value, that's as good as I would have dared hoped for.
The bad, for some reason it can only do that on low modes. I've done a ton of measurements. Off-time power consumption has nothing to do with what mode it was in before it turns off, because I always shut down PWM before sleeping. mcu power consumption while it detects shutdown does impact it though. LED consumption does not, because it's isolated by the diode from the MCU off-time cap, C2. If anything high LED current should drain C1 faster and result in quicker off-time detection before the mcu eats up its reserves.
Ok, so I measured mcu current draw in all modes (again). It doesn't vary much. With my powersaving trick enabled it's about 2.1mA for 1 7135 or the fet. It's closer to to 3mA for all 7135s (the difference actually being in-line with the spec sheet drive current for the 7135s, although it's interesting that these are technically still on when the FET is running, and yet the drain goes back down to 2.1 when the FET is on). As it turns out, I can get about 1.5 seconds of off time on the FET on full power if I add back a 500ohm bleeder, but I can barely get any off-time at all from the all 7135 channel. For the single 7135, again on low, it gets very long times. On full throttle at 3.0V it can squeeze out over 0.75 s, getting close to 2s by 3.5V, all with the bleeder, so bit worse than the FET. This seems to indicate it's not about total LED current. But I'm at a bit of a loss what it is about.
It's very possible that I just missed this in the first build. I was mostly testing on low or turbo, and didn't even have the backside 7135s installed on that build.
The extra 0.8 or so mA of the all 7135s does not seem to add up to anywhere near enough too explain this, from the math or experience (I did some work to get it down from 4mA in the first place, and yes it helped, but no where near this drastic). The math from all the measurements actually adds up pretty well in line with the 3s+ I'm getting on the low mode.
hmm...