Not sure if this is directly related, but i also installed nginx 0.8.54 /php-fpm 5.3.5 for comparisons, and it's even worse with nginx, test.php apachebench run with =>200 concurrency, 2nd run already drops from 5500 rps to consistent 330 rps.
Looked in my messages log and found heaps of logged entries for
tail -500 /var/log/messages
Jan 23 02:18:22 localhost kernel: printk: 206 messages suppressed.
Jan 23 02:18:22 localhost kernel: ip_conntrack: table full, dropping packet.
so raised value in /etc/sysctl.conf
no more ip_conntrack messages, but still subsequent apachebench runs back to back end with lower requests per second with test.php test, worse being nginx (can't handle =>200 concurrency), then apache (past >600 concurrency), while litespeed handles it better (up to 1000 concurrency).
maybe need some tuning http://timanovsky.wordpress.com/2009...-ip_conntrack/
To confirm the issue ran some static test.txt file tests and nginx was fine, while apache still exhibited the problem at higher concurrency levels http://www.litespeedtech.com/support...ead.php?t=4617
Edit: solved nginx php issue changed php-fpm from tcp to using unix socket