Coherence time of MCU class crystal ocsillator?

I'm guessing less than a minute, but that's based on just about nothing but a feel.

(And assuming that you've corrected out any frequency error; essentially that you've phase locked it, or phase locked to it, and are looking for the time to lose coherence after the reference signal is removed).

Cheap, but selected, XOs have cycle-to-cycle jitter below 10 ps RMS. If you use them to time out a period, and plot jitter against that time, the jitter stays there up to roughly a millisecond, then goes up linearly. There's a lot of variation among parts.

RMS jitter might be, say, 10 or 20 ns per second. So a decent 10 MHz XO might randomly walk 90 degrees in a second or three. Putting a cover over it, to keep air drafts off, can improve a cheap XO considerably. A good power supply is important, too.

The opposite extreme is a serious ovenized SC-cut oscillator, with jitter of a few ps RMS at one second.