Thanks Max,
so today I turned the rear motor pair about a degree anticlockwise. This has brought my mot_yaw down closer to 0 and evened up the work done by the motors. I'll turn it a little bit further tomorrow to get it right around 0. I also tidied up the wiring layout. There were a couple of torroids on the ESC cables resting on the shelf the AQ is mounted to, which could have potentially bounced and amplified any vibrations, so they've been secured.
I thought I would give Larry's idea a go, so changed my motor mix table so that the roll pairs of motors were both +/-50 instead of 100, and then went back to closer to standard PIDs. I kept the tilt rate D at 4500, but using 60 for tilt angle P oscillated quite a bit, but it feels quite good at 35. I reduced tilt angle D to 1644 and it improved further. Is power definitely limited by having lower mix values? I might try going back to the +/-100 mix tomorrow to see if balancing up the yaw hasn't helped with that too.
In the tilt test if I give it full roll stick, it will reach a precise angle (not sure how to tell how much?) and holds it without problem, but if I let go of the stick, it will oscillate slightly sometimes when it gets back to being flat. Sometimes if I threw it one way hard, then hard the opposite way it would almost look like it would overcorrect (but never crashed) Motors aren't clipping, but mot_pitch is closer to than mot_roll.
Some pics of the build - All the ESC's and high current stuff is down the bottom, the middle tray is pretty empty apart from a 12v buck converter for the LEDs and fpv cam, and my frsky rx. I wonder if the RX would be affecting the AQ board by being below it. It could be relocated. I have a 3DR radio on one of the other leg clamps, but I think everything's fairly well separated.