I don't thing that benchmark is valid, your are doing tests with an OS that is not compiled for athlon nor p4, and with no real multithread, so I think that is no valid. I'll buy a dual k7-xp@2000 soon, when I have it i'll write somethin' here... uh.. and with linux compildef for xp, that means REAL multithreading & binaries really optimized for the cpu's that I use...

and sorry for my poor english

Win XP Pro not SMP capable? Can somebody please call Bill at Microsoft I think we have a problem here!

You're right in that he's mostly a lunatic. But there's a little merit in the fact that when you compile a Linux kernel you tell it what CPU you're compiling for and it enables some CPU-specific optimizations. And the compiler is allowed to generate asm specifically for that CPU. Since all of them are x86 compatible and the kernel has a strict policy of not allowing fp math in kernel mode it's not enough to invalidate the test. But the P4 is notorious for needing special assembler to get good performance because of it's extremely long pipeline. And since XP is compiled for the lowest common demoninator (in this case probably Pentium w/o mmx) it can't be made to take advantage of that.

Also the apps used for benchmarking, c libraries and any other dynamically linked libs, would need recompiled for the specific CPU to get the best results.

Is it worth it? Probably not. Could it make a noticable difference, very possibly.