Two counters sharing a cache line across two NUMA nodes

Two counters share a single 64-byte cache line, exactly as in a same-socket false-sharing scenario, but now Thread 1 is pinned to a core on NUMA node 0 and Thread 2 is pinned to a core on NUMA node 1, and the cache line's backing physical memory is homed on node 0 (typical, since it was first touched by code running there under a first-touch allocation policy). How does pinning the two threads to different NUMA nodes change the false-sharing analysis, if at all?

31 of 47 in Computer Architecture