Hi Bill,
You may want to look into whether vMotion events occurred around the failure time. This vSphere feature 'freezes' the VM instance and moves it to a new physical host. During this time the VM host buffers incoming network packets, so clients would not immediately see this as a disconnect, but continue to send requests.
Setting up another network client to send ICMP pings to the VM guest once a second would be able to reveal if this issue is caused by network unavailability.
We've also seen similar issues in the past with overly aggressive VMWare traffic shaping.
Hope this helps,
Mikel
