Response times with and without varnish
We were recently asked to migrate a client running a Magento store on a complex infrastructure which utilised a varnish caching proxy, to a system that didn’t have varnish to reduce management costs (amongst other things).
The slowdown you can easily see in the response time from the server as reported by Pingdom:

Despite the drop in response time, the resulting high-powered non-varnished based standard apache delivered site actually ran very quick, and in some cases outperformed the SSD based server in index rebuilds for example – that’ll be the 6 virtual CPU’s and the 28Gb of RAM.