I'm able to reproduce this problem in the same environment. In about 5 of 10 reboots corosync start is successful, other ones end up with 7 corosync zomby processes and CRM which is not able to connect to the cluster.