I have made the problem a lot better by adding a 200uF cap across the output of the regulator. My theory is that keeps the PIC alive long enough to drive the FET completely off. The fault happens less often now, but it still happens.