Excuse me, but a 70% difference is hardly negligible.
For the creator of the compiler - yes, it is not negligible. It is probably a bug in one of the compilers (some optimisations not applied correctly).
For someone who decides which language to pick and for the final user of the product - except if it is an AAA game, it is negligible. Most of the C/C++ software out there could be significantly sped up by better algorithms, or hand coded assembly, etc. But almost nobody cares, except some CS students that have just picked C or C++ and want to show how smart they are. I've yet to see some software (in any language) that really uses all the power of my computer. What I really care for, is that there are no suboptimalities of magnitude 1000 times, e.g. like there are in the recent OpenOffice release (written in C++), or hangups like in recent Firefox (if the download list is too long - it freezes for seconds, ofc, also in C++), or crashes like in KDE, or delays like in Eclipse 3.6 (caused not by GC, but by sloppy coding). It just has to have acceptable performance, not maximum performance.
*Though I suspect you'd gladly accept such a benchmark if it gave Java the edge.
I've already shown the exactly same code running on Java@Core2Duo significantly faster.
But no, I would not accept is as evidence. It would show only that there is a still room for improvement in the compiler technology, on both sides, but particularly in VMs, which are younger than static compilers. But these are compiler implementations, not languages. Excelsior shows that Java can be compiled statically to as efficient code as C++ can be.
(Actually I wanted to say 64 bit Java on MacBook is performance-wise crappy and that is all - they have to work harder. But GCC several years ago was also crappy and produced extremely suboptimal machine code - yet no-one claimed they have to write in assembly instead of C++, because of that. See: http://stackoverflow.com/questions/1834607/64-bit-java-vm-runs-app-10x-slower).
At least simplas2002 is producing some empirical evidence.
The only evidence he produced is that some alpha version of a C++ compiler outperformed Java on some niche hardware platform. But on the most popular hardware/OS platform, providing his suboptimalities in code are corrected, there is a tie.
On the other hand, I can show you some niche hardware platforms, where you cannot write C++ code (because there is no native API and publicly availabale compiler), but Java runs perfectly. Would it be any evidence that "Java is better"? No.
So, basically what you're saying is that there's no benchmark that can be used as evidence that C++ is faster, correct? *
First define what does it mean "C++ is faster?". If it means that the best C++ compilers produce sometimes better code than some not-so-good Java VMs, then yes, I agree. However, the opposite is also true. Want to compare VC++ with Excelsior JET?