This explains why massively parallel HPC codes are mostly minimal mutable state designs despite seemingly poor theoretical properties for parallelism. Real world performance and scalability is dictated by minimization of memory copies and maximization of cache disjointness.