Continuos workload triggers 'unknown primitive cmd' message in PP log

When a customer workload is run using two node cluster there are sporadic 'unknown primitive cmd' error messages in PP log.
This tells that some ByteStream messages are mistreated, namely either ser part add extra to ByteStream or deser unexpectedly reads from BS.
I assume this failure can be seen in a multi-node cluster only b/c Allen couldn't repro this with a single-node cluster and similar workload.