The actual issue is parallelising it appropriately -- realistically the best you can do (with this approach) is hoist computation into a worker (which i'm working on) and you get hit by data transfer costs at that point as workers do not allow any form of data sharing between threads.

I'm not sure if there is any way to feasibly improve threading of computation in a non-shared-state threading model.