OK thanks for your help, so my summary:
(1) spreading merged tables across disks will only help concurrent
queries (in my data-warehouse application I'm doing things serially).
(2) there's no efficiency in the way a merge table splits the indexes
into smaller files -- if anything, it will be less efficient. *Unless*
you bypass the merge table and go directly to the underlying table,
which would require some client-side logic.
(3) mysql 5.1 has data partitioning, but that's at least 6 months away?
So: my original problem was trying to decide whether to split into
month-chunks or week-chunks. I think I will choose month-chunks, which
will make data management easier, and there are no performance gains
to be expected from going with week-chunks.