We set LOCAL_ONE for AQS in https://gerrit.wikimedia.org/r/#/c/267924/ as mitigation step to reduce latencies on the old cluster (no SSDs at the time). While this was a good idea at the time, it might be better now to restore LOCAL_QUORUM to leverage the read repairs consistency.
We'll pay some latency of course, but I don't expect it to be that much. Rolling back would be a matter of a puppet deploy in case the performance hit will be unacceptable.