The Java Faster than C++ Benchmark is bound to start a fight; anything that goes against the ingrained assumptions always does.
I am glad someone is still challenging those assumptions. I think everyone will admit that Java has come a long way recently and perhaps C++ has stood still for a while. In my own experience there have been some examples of very fast Java applications.
The problem with benchmarks is that they are not very representative of how fast your resulting application will be. As Keith will admit, this benchmark is produced by a Java expert rather than a C++ expert; and so I am not surprised that Java has done well. The real question is: Is your average bit of Java code faster than your average bit of C++ code? The answer to that question is one for the statisticians rather than the benchmarkers.
The fact of the matter is that the programming language is only a tiny factor in the actual performance of the application when considered as a direct influence, but when you consider the indirect influences, like do Java programmers have more experience in writing real-time software? then the language choice becomes significant - but impossibe to measure in advance.
Does anyone know of any statistical comparisons?