Copying is fast if the access pattern is a good fit for the CPU architecture. If you need to copy only every N-th byte from a AoS it might be as inefficient as

Copying is fast if the access pattern is a good fit for the CPU architecture. If you need to copy only every N-th byte from a AoS it might be as inefficient as random access. So, copying could be expensive. The article suggests striping your data in blocks but then you end up with the worst of both worlds in terms of program code complexity.