Take a closer look at VPSBible's results, their comments on possible issues of accuracy and their overall conclusion. Personally it's been my experience that php-fpm has been better suited to my particular use, but I've also never explored -cgi. Think of your current setup like this: You're running a benchmark you don't understand, that's spawning three thousand connection attempts per second, to a mis/poorly-configured server that is doing its best to keep up throwing error messages, to prove that one bad configuration withstands a prison beating better than the other. It looks like I get up to around 300 hits/second with your example PHP script on my Atom 230 with iSCSI storage, before changing that particular server's config for the task either way. (Edit: Oops - The errors were due to tcp conntrack limits for this kernel, then when I raised that, file handles hitting the php socket file. I'm not about to pick apart this system's config to see how high I can get that number, but I'm up over 500/second before other resource limitations start becoming an issue) And really, regardless of your opinion on what I'm saying or what tchen is saying, your results just don't stand up to any level of scrutiny. Learn how to use your benchmarking tool, bring your users down to a number your box can hope to handle, fix your configuration and then try again - at least you'd have meaningful data to present at that point. Your current results are useful for planning just how hosed you are under a DDOS, and little more.