I might not have enough context, but is it possible based on your architecture the shared queue could back up if specific destination endpoints aren't accessible when workers are processing the queue (I have seen this happen before in an adjacent space product)? If this isn't the case, tuning is probably not necessary if temporarily undeliverable messages can be skipped and retried later without impacting queue throughput. Being able to select, group and pause processing of queued messages in an ad hoc fashion (tuning was likely the wrong word, my apologies), I've found, is a powerful feature of webhook engines.