Allocations on multi threaded matrix operations (caused by BigFloats)

Perhaps @JeffreySarnoff knows some good alternatives?