Linux/Windows bitfile equivalence?

It doesn't use any random generator, all the "randomness" is generated from the seed. Each run with the same project files, run from scratch (i.e. without incremental compilation or anything taken from a previous compile), with a given Quartus version and platform will give the same result. Using different platforms (Windows / Linux, or 32-bit / 64-bits) will give different results. I'm not sure of the exact reason why the Linux and Windows versions give different results. The real question is why is it so important? Different seeds will also give you different results, but if your project is constrained correctly it should just work as well.