BrowserUK, an older post of yours on the subject (from 2006) enlightened me, and I was able to find out what the problem was and solved my issues. Thank you so much..

Gah! You're too generous. I should have remembered and recognised that problem and saved you a couple of days. I'm glad you've achieved your goal.

However, I still fail to see how spawning a new thread, or just transfering the socket to a thread pool is going to operate more quickly than just printing one short line, closing the connection and going back to the accept. Whilst a single push to a queue is going to be faster than IO to a socket, given that you're now setting up the listener queue, the tcp stack should be able to buffer the connections long enough to make the difference moot.

Did you do any soak testing yet?