Although you can use numexpr for some easy (albeit modest) threaded speedups and streamlined memory usage. I've had more success coding custom array processing routines in cython where I can easily exploit threads.