I had an issue where about 1 in about every 15 sub exposures would not guide well and oscillate a few arc seconds (RA or DEC). I was using PHD2 and dithering / pausing between exposures. It only occurred immediately after the dither and would continue for the full sub exposure. The next exposure, or a stop/start would fix the issue immediately. (Paramount MX, SGP, Pulseguide or ST4, any guiding algorithm)
I posted my issue and log files in the PHD2 forum and the development team compared what occured after a stop/start with pause and dithering. They did a brilliant job and introduced the same reset into the guiding variables after a dither. I have been evaluating a trial version which has performed consistently for 50 subs. The code should be making it into the next PHD2 release and make the PHD2/SGP combination even better.