avoid future implementation mischief. (Maybe not. Sometimes, general distrust of theory leads people to distrust simple reasoning over times from CPUs trying as hard as possible to mask DRAM latency via pre-fetch.)
..<
.. ^