> Do the other readers of this list have any memorable/interesting 
> papers to cite?

No papers, but it made a massive difference to the performance of some
big systems.  Especially those where cache snooping was slow.

As well as process affinities if can be of great benefit to have per
cpu free lists for memory items and to keep both ends of an active
pipe on the same cpu.

Of course certain workloads will pump memory resources from one cpu to
another - unless you allow for this you get unexpected resource crisis.

	David

-- 
David Laight: david@l8s.co.uk