I have a 3 host cluster setup with HA enabled and fencing enabled and it appears to be working properly. Executing power management stop, start, and restart work along with host shutdown/restart following a simulated crash. When network is pulled a proxy is chosen and it powers off the downed host, and then restarts it. Since the network is still down it repeats the following in events:
"Host kvm01 is not responding. It will stay in Connecting state for a grace period of 162 seconds and after that an attempt to fence the host will be issued."

The real problem here is that the VM's on the host that has failed never migrate to a new host and remain down until the network is reconnected.

We have tested this with back-end storage on gluster and NFS with the same result. This is on oVirt Engine Version: 3.5.1.1-1.el6. Hosts are on CentOS 7 and the Engine is standalone on CentOS 6.6.