My question was rhetorical. I think we're in agreement - memory accesses may require page-in which requires OS scheduling - but to call the latency of a memory access a type of blocking is sort of confusing/misleading. It may take a long time, and you may get preempted and never return, but those are sorts of different things.