Cracked the problem(s).
I'm currently letting the PHD2 Guiding Assistant run as I post. It's gone through nearly 2 cycles and my peak to peak is now 11.7. (Conditions here are 20k winds outside obs; skippy sky thinks seeing is going to be a '5'; it's cloud free as far as I can see; I think the seeing is spiking around 3+")
So how did I fix it? Well, the regreasing, etc has certainly helped and I might be able to reduce that number further with some more work in that area.
However the key fix was improving the PA further. You might be thinking wtf, didn't I say it was within 1'? Yes, I did - I unwittingly was telling a Big Lie.
Here's the first lesson: don't believe the PA error reported by the guiding assistant. It's only loosely based on a true story.
When I started tonight, I wanted to capture a baseline before making any other changes, so I thought, go back to basics and check alt/az. I started with alt tonight as I was wanting to do final testing in sky where I would be adjusting the azimuth.
I set up the drift align tool for alt and then get a massive WTF?? From memory it was out by about 13'. As I went to make the adjustment I noticed the locking bolt was not as tight as it could be. I'm not sure why it's that way but I'm getting some suspicions that it's not locking as well as it should and slips over time.
I fix that up and then adjust the azimuth. It was pretty close, but I spent the effort to refine it further. As I was doing it, I noticed that the wilder fluctuations were disappearing.
Finally, I went back to the Guiding Assistant to find the peak to peak value and to see how the graph appeared. It was much shallower of course and didn't have too much rapid changes on the rise. The seeing spikes were more observable (to me at least I think).
I've switched off the assistant and am now guding. The "dart board" is getting better grouping. See the pic below (note the 'y' scale is +/-4"):
So, stoked to be back in the game again. Thanks everyone for the suggestions.