Have you found source of this behaviour ? We’re experiencing the same issue for smaller machines 8GB and 16 GB. VMotion process takes quite long time but network infrastructure doesn’t seem to be the culprit. All the VMotion traffic is closed within Blade chassis (HP Virtual Connect) with 1GB internal links. The traffic seems to freeze for several second, then jumps to 100MBps and dies again.
Maybe it is buggy ESXi – 4.1u1. Two problematic clusters are built on this version. Another two are 4.1 and the issue doesn't exist.