Large variations between sub exposures can be a result of a seeing/scintillation affecting the guiding. (The star is jumping in and out of the slit). Guiding on very bright stars can be problematic. You need a well exposed (ie not saturated) star image to guide on but to avoid saturation the exposures end up being very short. The guider then makes too frequent adjustments on seeing variations. PHD is also prone to hunting about the slit position when guiding using the overspill from the slit. (PHD tends to lock onto the brightest half of the split star image rather than the centre of the star) Using one of the alternative algorithms in PHD eg hysteresis or slow pass filter can sometimes help in these cases.