I believe that all gurus here (that are usually respond in this forum) are running not released software that was improved a lot (thank you guys - you are awesome!), and are not really aware of existence of such problems (it may happen that this problem exist only with some clones of revo boards and 15.09 sw).
My advice to try LP next before you switch to different FC. (It is really awesome!)
It's in human nature to look for patterns, but often people draw mistakenly conclusions from those patterns trying to explain something, while in fact the real cause is completely different. Same happens with changing firmware, or hardware to completely new one. You change everything so it's likely you changed something on the way you don't know about. I am not excluding any possibility of the cause, but saying it may not necessarily be firmware problem.
Does going back to the old firmware with the same procedure/similar config causes issues ?
Another thing is that LP uses 3D magnetometer. This usually causes most headache to people. What that means is that LP measures magnetic field at all
three x,y and z axis and uses all that information. While some other software may use only 2D magnetometer and use only two axis, ignoring third measurement. Having more data helps EKF to give more accurate estimates, than just using fewer data. However, despite the EKF (aka INS31) is designed to deal with noise by taking best of all sensors and blending them, it can not handle completely noisy, contradicting or not consistent data.
Often people fly in Complementary and think that enabling INS31 just adds GPS. That is not correct. Complementary uses Accel+Gyro, while INS31 uses
all sensors available on the aircraft.
If any of those sensors produces totally wrong data either because of noise or bad mag rotation (in case of auxmag) that are not in agreement with the rest of sensors, then fusion algorithm won't be able to give reliable results.