That's my question too... There's a reason there's such a thing as a cache hierarchy.

The address space may be flat, but the latency will vary a lot depending on what part of the memory you're dealing with.

https://en.wikipedia.org/wiki/NUMAlink