Replacement of the crystal has eliminated the instability, and with a minor adjustment of the pause values I have a sufficiently accurate cycle time for the application, so it looks like a dodgy oscillator was the cause of the problem.

Many thanks for all the advice - it's been an education.