A more realistic test would have to be much more complex and test multiple areas of the language.
There is a "Sieve of Eratosthenes" reference implementation (as-in all the different versions do the same amount of work, disregarding specific language optimizations, relying only on compiler performance) in pretty much every language you care to test on for just this kind of speed comparison. Some may disagree with me on this, but I've found that good performance isn't really measured in clock cycles. It's measured in "acceptable usability". What defines acceptable usability differs from application to application.
Well said.