Paul -

Speaking from experience running HSM on AIX JFS for many years:
Applied to a file system, HSM turns it into the slowest you've ever
experienced, even on the fastest RS/6000 and disk systems.
Traversing it for an incremental backup is very painful.  Sizing an
HSM file system is a challenge: on the one hand you need a big enough
"window" for the largest file to fit through, but on the other hand
you want utilization moderate to avoid congestion.  With too many
utilized inodes, it's all the scout daemon can do to refresh the
migration candidates list in time to make space for new files.  As a
gross, simplistic number, I would advise limiting such a file system
to about 500K inodes used.

   Richard Sims   at Boston University