Based on the manual, and some comments on CN, it appears that currently, SGP allows PHD five minutes to recover guiding after it loses a guide star before going into recovery mode. That can be problematic for small FOV scopes because if a cloud goes by, depending on how close the stars are, how much drift the mount has, and how long it has been, PHD can end up selecting a different star and the centering can be off. One option to address this would simply to be the user to set a shorter timeout before recovery mode kicks in. If you know your gear is subject to some drift, you might set a 1 minute timeout for example. Once recovery mode kicks in, then a plate solve would occur to correct for any centering issues. Another option would be to have a checkbox to force a plate solve whenever the guide star is lost and then returns.
Interesting. I believe I had this same situation occur a few weeks ago. The last 15 subs were all shifted in RA by about 1/4 degree.