I am also struggling to know the reason for several crashed with two planes, so I share your pain
I am just starting to learn how to analyze the logs, but I see something that could be related. for plane 1, if you plot the “mechanical failure” during FbwA, there is a huge offset between desired and real roll and pitch, in my logs when a plane is flying ok, they have similar values and trend. for example, in your log for plane 1:
roll is mostly negative but demand is positive. Looking at the parameters, RLL2SRV_P and PTCH2SRV_P are both 0.4, which seems to me a very low value (it is around 1 for most models, I think). For X8, which is a simmilar airframe , RLL2SRV_P is recommended at 0.8 (http://ardupilot.org/plane/docs/configuration-files-for-common-airframes.html)
Maybe the P value is too low and the I value is not enough to reach the demanded bank in a reasonable time?
Did you tried autotune?