[saltcured]

The caching effects mentioned there have to do with adding additional hardware resources that change the effective cache sizes and shift the working set into higher performing storage or memory.

You can also encounter a funny software engineering effect. Refactoring an array-based algorithm to work in parallel often involves the introduction of block-decomposition. This change allows sub-problems to be spread to different workers. Sometimes, this change will also accelerate the sequential code, as the new block-oriented access patterns improve cache efficiency!