So I left it where I was and was able to improve things by first binning the guide camera frames 2x2, some noise reduction, and I was able to get by with the 1.5 second duration I was using previously. It did require creating a new profile in PHD2, and a re-calibration, a new bad pixel map, and for whatever reason a new drift alignment.
Sounds like a lot and frankly I forgot how to do much of that but all in all only about an hour or less (not including the pixel map which I shot before dark). Soon enough I was back to capturing frames with decent but not great guiding.
The previous nights sequence ran fine until the last object failed to autocenter. I forgot to look into that so it probably failed again last night