If running walks, each profile should only run on one machine or the other at once. If you have a very large profile, it's recommended to split it into two smaller profiles, and then index one profile on one machine and one profile on the other, letting them replicate to each other.
Using circular replication behind a load balancer can make troubleshooting issues more difficult versus a standard "sender and receiver", as it's often unclear which one sent or received an individual piece of data.