I haven't encountered a case where multithreading will make the algorithm weaker, but I do have a variation of the benchmark code(on disk, at the moment) that will spin up multiple worker threads to compute PoW.