Hi Shaun,
Not wanting to turn this into an Oracle list but.... count(rowid) is not
measurably different from count(*) - or count(1).... the problem with these
closed source databases is that we have no idea what the code is doing ( e.g
: is count(*) resolved into count(rowid) , count(1) or something else for
that matter ?), leaving you with the only option of measuring the various
possibilities....and trying to account for your measurement error.
I get - using the current big O release ( 9.0.1 ) on the infamous fact0 table
38 +- 1 s for the lot.
My 2c worth is that this issue consumes too much intellectual energy for very
little gain ( 1/38 x 100 % possible gain at the best.)... there are usually
more productive things to tune for orders of magnitude of reward....
regards and best wishes
Mark