In short, you are pushing messages to one other cluster node at a higher rate than serializing and passing messages it across the network can keep up with. If it is just temporary bursts, increasing the queue size can help, but that will not help for scenarios where you constantly push more messages than can be handled. If serialization is the bottle neck, and the recipients are different actors, tuning outbound-lanes can help.
The problem can also possibly be caused by head of line blocking, if a few of the messages are very large so take more time to serialize and send, that can make other smaller messages queue up behind it. Isolating larger messages to their own message channel can help if that is the problem: Artery Remoting • Akka Documentation