When Casper wants to stream a block to a remote peer and the outgoing message queue is full then comm should drop the block. Currently, Casper waits (forever) until it can enqueue the block.
When streaming of a block to a remote peer fails then don't retry. Currently, we have 3 retries with a 1-second delay.
Make the outgoing and the incoming message queues much shorter (f.e. 100). Currently, we allow a node to fall 1000 messages (blocks) behind before we start dropping messages.