In the proxy server that I wrote back in 2000's, relying on the Windows NT/2000 scheduler for multiple CPUs was a big performance hit.

Threads would jump around cores, killing most performance gains by using multiple threads.

The workaround was the typical getting the amount of CPUs and park each thread on their own one, this alone made quite an observable improvement.

I think by Windows Server 2003, this scheduler issue was already sorted out.