If your polar alignment is good, then the guide star should not drift significantly in 10 minutes. What is your measured polar alignment error?
The most likely way for phd2 to pull you off target when clouds move in is for it to lock on to a hot pixel when the clouds are present. Are you using darks or a bad-pixel map in PHD2? A good bad-pixel map can effectively eliminate this problem. Your large PHD2 search region (80 pixels) may be exacerbating the problem by increasing the chance of locking on to a hot pixel.
If you post your PHD2 debug log and guide log from the session we may be able to say more.
Andy