Hi,
Beware that if using a single vCenter Server for the two sites, you need to have it up and running, in order to perform failover of a replicated VM.
The suggested topology in vSphere Replication documentation is for multiple remote sites replicating to a single DR site and vCenter Server and VR appliance running in that master site. Should the vCenter Server fail, VR can not be used to assemble consistent VM image from the replicated files.
About replication traffic traveling between the sites and back - the replication traffic goes from the source ESXi to the VR server and then to ESXi that has access to the target datastore.
If using SRM 5.0.x or 5.1.x with vSphere Replication, you can install multiple additional VR servers and that way the replication traffic does not need to travel between the sites and back.
In the upcoming 5.5 release, even without SRM, in the Web Client UI, you can install multiple additional VR servers.
After the initial full-sync completes, the bandwidth consumption is proportional to the changed data blocks at the source VM. If in a given RPO interval the same disk block is changed multiple times, it will be transferred only once. As commented above, you might need to adjust the RPO value.
Regards,
Martin