I would caution against speculating about TSM database cache
algorithms and BUFPoolsize value effects without supporting
evidence.  I would not expect its use of database records in memory
to be substantially different than its indexing of records on disk,
where I doubt that there would be a lot of "searching through memory".

Performance degradation with elevated BUFPoolsize levels is
traditionally due to paging, as the TSM Performance Tuning Guide
explains in its recommendations for values.  (There are supplementary
Technotes.)

Operating system memory utilization is always a complex topic, which
also changes as each OS evolves.  Having a large amount of memory is
wonderful; but one can't assume that the OS is going to apportion its
utilization as we might expect.  An OS typically computes sizes for
various reserved areas at boot time, according to the ingredients
defined to the system, and hardware discovered.  The amount of memory
left for application use may be substantially less than total
memory.  Environmental values under which a server process is started
can also limit how much memory it is allowed to work with.

So, be sure to pursue analysis where you perceive performance
characteristics you don't expect.  Enlist your OS specialist as
necessary.

   Richard Sims   at Boston University