I suspect it's a combination of the R/C network and the ceramic oscillator - if the PIC runs slightly slow, and/or the R/C time constant is slightly too small the problem is seen. If *either* the PIC oscillator runs a bit quicker or the R/C constant is larger (bigger R and/or bigger C) then the problem is not seen.

The capacitor fix attacks the output of the monostable as opposed to the actual timing - the glitches still happen but the capacitor hides them. But... the fix can be done without removing the display.

I'm glad that the fix appears to have worked for everyone, pity I didn't find the glitch about a year earlier! One issue was having a unit which exhibited the problem 100% of the time (hot or not) in my hands - when I had this it became a lot easier!

Hugo