A very long time ago (1994), one Jeff Tranter wrote a bogomips calculation program, based on the current linux kernel code at the time. However, since then, the kernel has become far more complex and supports many more architectures. While the basic bogomips calculation code hasn't changed since that time, each platform has its own version of __delay(), some of which result in higher values than Jeff's tight for loop. Consequently, I've fished the various versions of __delay out of kernel 2.4.16, and modified the system to use these instead.
Download the source here: bogomips-1.4.1.tar.gz (5K).
Download a Windows binary here (386 delay): bogomips.exe (52K)