@Bill:
I´ve flown using TILT_ANGLE_I = 0 (equivalent than limiting IM) without any significant result, when forcing the divergence.
It should be nice to log the evolution of the P,I,D terms for both rate and angle tilt, in order to analyce them later. I think there is the key to solve the problem. We´ve done that in Arducopter loging and even printing and plotting P-I-D terms in rear time to tune the Loiter (we had problems with PID´s there) and was very useful to solve the problem.
Angel