The performance comparison of C#, Java and C++ was always a controversial subject. Many people who are unfamiliar with JIT'ed environments claimed anything runs on JIT is too slow compared to C++. On the other hand people who actually use Java and C# usually claimed the otherwise that performance impact is negligable. Here's a nice chart I've found that a friend in freenode IRC server pointed me. These benchmarks are done by Matt R Warren (aka Moridin8 in irc.freenode.net) and his original site is http://www.csharp-architect.com
According to below chart, choosing C++ just for the sake of performance doesn't seem very feasable to me. Remember : Engineers are expensive and servers are not!