Most appreciated, gentlemen! I'll give it a try in the coming weeks and will report back! A couple more thoughts below.
if the transient is not too bad, it should work.
i would, however, split the rail: using the 2.7v to power the mcu and the dc/dc converter for the 5v device.
The datasheet for the boost regulator has a scope trace showing the transient, and it seems decently smooth. I thought about splitting the rail, but then I may have to level shift signals between the MCU and sensors, I'd prefer to avoid that if possible.
I would think it should work ok. The only caveat I can think of is that the brown out voltage detector reset might have to be disabled or set to a value lower then 2.7?
I said 2.7V because that's the BOD setting I'd use. The 1.8V setting seems too low to me. An in-between setting might be nice, but such is life! Certainly the battery will have some juice left at 2.7V, so it may be worth trying 1.8V. In that case I'd want to slow the system clock before turning the boost off, but that's straightforward enough.
I can't think of anything in the datasheet that contradicts what you want to do. Seems like a good idea to me.
I did find an Atmel application note about using the BOD that talked about variations in supply voltage, and the typical problems if they got below recommended levels, this is why the BOD should be used, etc. So that would seem to imply some variation is tolerable, but it didn't go into detail on that specifically.
If the processor will spend anytime polling 1 MHz (or even 128 KHz) may be a better choice.
Probably not much if any, but are you thinking more from a power conservation standpoint, or immunity to the Vcc fluctuation?