I did something very similar on a 5 read cluster, but wrote custom NetScaler health checks to pull laggy replicas out of the virtual server pool.