Hovering the Quad' was OK, but it still seemed to be unstable if a gust hit it or too much control was put in. This suggested the P term in the PID tuning was too large, causing oscillations if the difference between the commanded stick input and the measured attitude of the 'copter was too large. So I used the quadcopter on strings method to tune out P and I for pitch and roll (and had a stab at the yaw values). The results seemed good, so I found a nice large field in Devon, grabbed three battery packs and went flying...
PID values used were (low pass filter on the MPU-6050 set to 42Hz):
Roll and pitch: P=1.2 I=0.008 D=3
Yaw: P=2.5 I=0.016 D=0
Next step is to calibrate the auto-level!
Oh, and the crash at the end resulted in a snapped prop, and an excuse to buy lurid green and orange props :)