Regarding your Revo problem.
The LibrePilot software uses 3D magnetometer (all three axis of space), in contrast APM uses only 2D magnetometer.
It makes no difference with Complementary, but EKF benefits greatly from getting more data as it makes estimates more accurate. On another hand, less data as in 2D case, makes algorithm more robust to noise.
If you don't include noisy measurements, your estimates won't be affected. The EKF is designed to deal with noise but to certain level, it just mixes best of all sensors. If one of your sensors gives total crap it just wont make it better. As people hinted on forum, your nu-metal may distort subtle magnetic earth field, your internal magnetometer might be affected by switching components or power wires.
The solution for this is either not to use EKF, or use external magnetometer, twisted wires if necessary, proper outdoor calibration and no nu-metal thingy, that may distort subtle earth field.