I wonder about how eye contact will function in VR - cameras inside the goggles tracking eye direction? I suspect that eye contact is the kind of problem without a lot of room for error. Get it only slightly wrong and the effect could be very disconcerting.

I'd bet on cartoonish, stylized avatars being the most successful (at least initially). Cartoons work for a reason, we're very good at taking a cue and filling in the blanks. But the more information an image carries the deeper you go into the uncanny valley. For realistic VR avatars to catch on, they'd need to be nearly perfect.