Informix 7.12/SCO 5.0 Replication Problem

Scenario: The secondary server is disconnected from the network. The engine on the secondary server recognizes that replication has failed and turns replication off (onstat -g dri and onstat -m on the secondary reveals this). However, the primary thinks everything is OK. The primary is locked to all transactions (replication buffer on the primary is locked from a previous transaction). Simple selects work fine, however, anything involving writing to the logical logs freeze. It remains in this state for approximately 2 hours. We put a sniffer between the boxes and we discovered that TCP/IP continues to retransmit failed packets during the 2 hours. We found a parameter in /etc/default/inet that seemd to be the culprit. It was set to (keepalive parameter) 7200 seconds. We reduced this value. Nothing changed. Is there anywhere in SCO TCP/IP where I can set a parameter that will allow TCP to determine that the connection is history in a time frame quicker than two hours? Should the primary engine take its own initiative and not wait for TCP/IP to make this discovery? Help?