Couple of things:
1. Try running a traceflow between the vnics of the two VMs involved. This should show you where the breakdown is happening. It might be a DFW rule or VTEP can't talk for some reason.
2. Its possible that the netcpa service is hung on the host. This is the process that talks to the controllers. SSH to the hosts involved and issue the following command:
/etc/init.d/netcpad restart
