Table reads are repeated in proxy_cluster_pre_request, which can degrade performance. This becomes more noticeable as balancer members and client requests increase. Avoiding those repeat table reads here shows ~30%-50% performance improvement.