In order to have the best possible performance for the HA Load Balancer, it is necessary to have a round-robin first over the HAJournalServer instances using a light weight service such as HAProxy. That round-robin will distribute the queries without regard to the actual server load. The HAJournalServer instances will then load balance the queries according to the actual load and will use a threshold to prefer to handle queries locally. Effectively, this means that they only proxy away queries to correct a load imbalance rather than proxying 1/3rd of the queries (for an HA3 setup). This is a net win because HAProxy does less work than jetty to proxy the original request. This is also a win because the HAProxy will automatically monitor the health of the services and thus can hide service failures more effectively from clients.
Document this setup on the HALoadBalancer page of the wiki.
Verify performance on an HA3 cluster.