Upgrade to 6.8
If you read the changelog, or perhaps it was just in the thread for that release, I added that "error" message to indicate situations where a race condition was happening. It's not a real error though. I believe the behavior in the code is absolutely correct according to the windows socket documentation, however it clearly seemed to not be happy and to lockup from time to time. I made a number of tweaks and I think I resolved the problem.
I think this is further proof that the windows socket library has problems. I presume all the socket overhead or cpu load of an active IIS server is making the race condition more likely...
Try 6.8 and see how it goes.