vR Ops HA isn't meant or deisgned to handle spanning datacenters. The latency requirements are very tight, so spanning DCs makes it unfeasible at this time (v6.0.2). HA mode is meant to protect your data/cluster from failing if you lose a single node in the cluster. The recommendation is to put the replica on different host/datastore, so even if you leave the underlying resources for the master, you've still got you cluster config/master role online. I second your hope that HA will get better, and I'm certain that it will!