Perhaps the performance of your storage or network isn't up to scratch. I have a 3 node cluster with ceph on ssds and a 10gb network and my live migrations worked perfectly fine.
I've not seen flaky comms between virtual interfaces on this setup before, but you never know. ;)