ROSS automatically rounds up the size of all messages to either g_tw_event_msg_sz in the normal case or 500 bytes if ROSS_MEMORY is set. We could improve performance for models using multiple LP types by only transmitting the size of each event and handling short receives on the receiving side.