It isn't prop load or the epoxy - the shaft flexes at certain RPM. In rear-mounted motors, the motion is not adequately damped out by the bearings and this can allow the magnets to contact the stator. It physically just knocks them loose. It is worse on motors that are 'long and skinny' than those that are 'short and tubby'. Stunt flyers find it manifesting itself in tight turns, when the model rotates the motor mount/firewall, and the prop tries to keep going straight. I think that most of the stunt flyers went to front mounting, and were able to dispense with the extended shafts and external ball bearings that were an attempt to overcome this same problem.
When we were going through this with Scorpion, I sent my first motor back, they re-glued and tested 'at 2000 Watts' (not sure what kind of test they mean) and the magnets didn't come loose. On the first run with a propellor - I think it was a wood, 11-7 pusher, when I throttled back to idle and then up again, a noticeable sound (like a rattle or scraping) was made. I immediately shut down and landed and several of the magnets were visible in the opening between the bell and the base plate - indicating that they had slid rearwards. Sent it back again with a couple of diagrams of what I thought might be the problem, that the rear-mounting allows more shaft whip than front mounting. They recalled all of that model motor, and redesigned it with a large ball bearing supporting the rear of the housing. That fixed the problem. When I got my new-model motor back, I also redid the front end so that the motor was front mounted and it never made any more noise.
I had posted those drawings on Stunthanger, in the E-power forum, but I think that those postings were lost in the 'Big Crash'.
(edit - found a copy of the drawings)