@skiwi: Real-life example: I've got a C++ code base of ~600kloc here that's compiled nightly on three different architectures and compilers.
Windows 7 64bit, MSVC 10, Xeon E3-1230 @ 3.3 GHz, using 4 threads -- 3m 24s
Linux Mint 16 64bit, GCC 4.8.1, Xeon E5-1660 @ 3.3 GHz, using 8 threads -- 1m 42s
AIX 6.1 64bit, IBM XL 11.1, POWER @ 3.7 GHz, using 2 threads - 21m 18s