Here is another possible experiment which could indicate definitively whether the inductance of the outer bearing race is playing a role. Unfortunately it also seems likely to be harder than it appears at first.
Run the experiment with stainless (non-magnetic) bearings, and (hopefully!) observe that the motor doesn't go. Next, wrap many turns of uninsulated iron wire (obtained where, exactly? um) around the outside of each bearing race. The tricky bit is that the wire must make good electrical contact with the bearing race all the way around, because the idea is to increase the effective permeability of the race without messing up the current pattern in it. An obvious way to make sure everything's in good contact would be to *solder* the wire in place all the way around, but soldering to stainless steel isn't something I've ever done and I don't have the impression it's straightforward. Anyhow, if one can get the wire stuck to the bearings at all, then one should wind *enough* wire so that the magnetic properties of the wire are likely to make a gross change in the behavior of the outer race. Dunno how much that might be -- presumably, one would want the layer of wire to be at least as thick as the bearing race. Now, re-run the experiment. If it still doesn't go, then we haven't learned anything, because there could be any number of reasons why winding iron wire around the outside isn't the same as making the bearings from iron to start with. But, if the motor *does* go, then we will have demonstrated that it is the magnetic property of the iron, not its thermal properties, that is having the effect, /and/ we will have demonstrated that it's the magnetic properties of the *outer* *race*, rather than the balls or the inner race, which is important. Whatever; this is another one which would be time consuming and might or might not show anything...
