Benchmarking [CLP(B) libraries and other stuff]

No. That is also why I prefer a chart rather than a table. It gives a rough idea. Different hardware, different compilers, different OS, etc. cause significant differences. Even different runs easily make up to 10% difference. Also the individual benchmarks may be misleading. One might be tempted to conclude that clp(fd) of SWI-Prolog is nearly as fast as SICStus, but that is AFAIK pretty wrong. It is probably just a lucky hit in the ordering of the propagation for this specific test case.

It is nearly impossible to compare all that fairly. AFAIK, MinGW (GCC) by default does not optimize specifically for the target machine. The releases are compiled on AMD. The nightly builds for Windows on Intel, using otherwise the same Docker container.

Oh, the OS doesn’t matter too much. The Windows builds perform nearly equal to the Linux build (Windows build executed using Wine under Linux). Memory management, thread synchronization and file access are a lot slower in the Windows version, but none of the tests stress that.