My vCD deployments have both Console Proxy and HTTP on the same subnet (/24). If they were on different subnets, that would be fine too ... so long as each user can reach the address in the 'Public Addresses' section ... or the IP/Hostname if the value is blank in Public Addresses.
As soon as the cell is online 100% as per cell.log and you can login to the interface, you should be able to get a remote console. I've had a number of deployments work this way, including my two labs.
Are you using a load balancer? It's possible that the load balancer might not be correctly detecting that the virtual server in the LB Pool for Console Proxy is not up ... so it might take some time for it to detect and start routing traffic again. that'd be an odd ball, but not outside the realm of possibility.
Do you have a client system on the same subnet, or a way of running the VMRC directly on the RHEL/CentOS cell to test without networking even involved?