Thanks for all the thoughtful replies!
John, that's actually pretty accurate, we need a 1MHz bandwidth or so on that signal.
Tjarco, that makes sense. We can perform that test to see if the stack is indeed overflowing. As I said, the only reason we re-enable INT1 is because we still need to allow ADCINT1 to interrupt TINT0. I suppose we could accomplish that by using ADCINT1 in group 10, however (although it would be nicer if TINT0 was available in another group so it would by default have a lower priority than ADCINT1). As you said, though, we could also consider a different timer source (most likely a spare PWM, although we're running low on those, too).
Even if this is a stack overflow, I'm still not completely understanding why overloading a single interrupt (HRCAP2) could cause this overflow, since we never EINT within HRCAP2, so it should never become re-entrant or anything. Perhaps there's another interrupt causing a problem when HRCAP2 is running so often due to the overloaded input.
- James