Every platform has a limit on how frequently a cache line can be
atomically updated. In the general case with many CPUs updating
the cache line, there is an upper bound which is the reciprocal
of the cache-to-cache transfer latency for a dirty cache line.