Hi Shil -
I am not familiar enough with the Rackspace cloud platform to know for sure, but it's likely that the routing between your cluster members uses the WAN (public) network, rather than the LAN (private/local) network. Perhaps the hosts have multiple interfaces, which could allow private (LAN-based) routing between your cluster members? You might ask your tech rep at Rackspace to see if that is an option with their service.
However, per the Hazelcast docs, I believe you will need to secure an enterprise HZ license to enable WAN-based replication within your cluster. In that case, you may also be able to receive additional support building an appropriate Hazelcast configuration for the plugin, including an appropriate license key. The clustered Openfire deployments I have done to this point have been on private/local networks, and so do not typically encounter the timeout issues you are observing.
Regards,
Tom