causal_reads=local sends redundant MASTER_GTID_WAIT queries

If the GTID was already synchronized once on a backend, there's no need to synchronize it again. For workloads that do two or more reads after a write would benefit from tracking the last synchronized GTID.