Load Balancing based off TCP Payload entity

the entity ID will only be available in the first few TCP packets then it will no longer appear.

That could potentially be a problem. You'll need some piece of information from the client in each new TCP session after the initial load balancing decision to maintain that persistence. Is this a protocol that maintains one long TCP session, or is it like HTTP and can have multiple TCP sessions over the life of the application session?

If the latter, is there anything else unique about the client that can be used, or is supported by the client for persistence? Is source address an option?