In my experience with light dimmers, there's a few things I've come across with zero crossing detection:

phase shift due to voltage fluctuations
noise
assymetry

Voltage fluctuations causing phase shift most are no too critical - best to quantify the effect and decide if it is a problem.

Noise - always a problem. You need to filter the signal quite a bit to eliminate the garbage.

Assymetry. To get around this problem, I only detect zero crossing on half of the phase and use this to correct a software phase locked loop. As part of this you can also digitally filter the zero cross signal and correct for phase shift.

So at a pinch, you can get good performance from an opto on the mains side and not worry too much about the comparator woes you've described. To test for the effect voltage fluctuations have on the zero crossing, I would feed mains through a variac then into my circuit, have my micro output its version of the mains cycle as a square wave - feed the mains and this signal into the input channels of an oscilloscope. Have the channel with the mains signal overdriven so you see the zero crossing as a steep slope on the CRO and measure the time different between the two signals. Vary the variac to see the change. Wave a hot air gun over the circuit to see if temperature affects things, then try the freeze spray. Not entirely scientific but will give you an idea if voltage and/or temperature causes drift.