Don't use auto, set worker_processes to a value where high performance keeps cpu(s) around 50% max, this can be 2, 4, etc...

Windows server 2012 is AFAIK the only version which can scale cpu's where the auto value has any use.

Some cpu's can easily handle 2-4 workers per cpu, some only 1 per cpu, this varies because there are a number of variables like cpu type, bus type and speed, xeon quads, how many real cpu lines going to each cpu, type of HT, etc....

We runs test with 4 workers on 1 vcpu which works fine on xen, hv and vb but not on vmware (2 max).

---
nginx for Windows http://nginx-win.ecsds.eu/