You might go back and re-read some of the material that was already posted in this thread and review what you have discovered:
* 5 second delay between the server receiving 8 ACK packets and the server sending out the next packet - this 5 second delay apparently only happens when the client is connected over a WAN connection with a 80ms to 100ms ping time (can you confirm this with a local client and Wireshark running on the server).
* The client ACKing after every 2 packets, while the server is apparently expecting an ACK after every 16 packets.
* The server and switches had mis-matched speed and duplex settings (one side set to Auto while the other was set to a specific speed) - is it possible that the two sides are still manually set to different configurations?
* Reviewing the network switch's logs should be able to tell you if the server is occasionally dropping off the network, or if there are packet framing problems.
If you enable a 10046 trace (at level 8 or 12) as well as a Wireshark capture on the server, are you able to locate the 5 second delay in the raw 10046 trace that corresponds to the 5 seconds of no network traffic? That might tell us what the database server thinks is happening at that point.
One other thought. How many network cables connect the server to the switch? If there are 2 or more network cables, did someone tell Windows to bridge the network cards (DO NOT DO THIS - it will create a network loop if both cards have IP addresses in the same subnet). If there are two or more cables, are the switch and server configured to bond the network connections into a single 200Mb/s (for two network cards) pipe - this requires configuration on both the swicth and the server?
Co-author of "Expert Oracle Practices: Oracle Database Administration from the Oak Table"
IT Manager/Oracle DBA
K&M Machine-Fabricating, Inc.