The riak_kv_replrtq_snk worker_count is intended to help control the pace at which replication references are fetched by a sink cluster. However, reducing the sink worker count can lead to a situation where fetching is no longer limited by the worker count, but instead by the per_peer_limit * count(peers).

This is a problem caused by the deferred_queue_length not being reset not being reset when the worker_count is updated.