changing sending ports in the middle of a job

A mainframe host is sending from a specific port to a PC (an IP
address / port) using "Raw" TCP/IP. At the end of the first page they
change from the original sending port to a different port, still
sending to the same IP address and destination port. The job fails,
only the first page printing.
I think it its correct behavior on the part of the receiving system to
end the TCP stream when transmission ceases from the original port,
and assume that the transmission from the new port (same IP address)
is a new transmission.

Am I incorrect?

Any ideas or suggestions?