Computer Language Shootout has some interesting figures. Note that Java does well, even compared to compiled C and C++. I'm sure C# would do better if the
test systems were Windows rather than Debian.
Keep in mind these benchmarks are for computationally intense tasks. For real-world applications that are bound by IO instead of CPU, I'd expect the performance to be far more similar among languages.
As Fred mentioned, one possible variable is going to be the programmer. These benchmarks test a number of solutions by a variety of programmers, so you can see this effect in the range of results for a particular benchmark and language.