Is that reasoning off-base, or is the right path to split these out per subsystem
regardless?

Adding to what others have said, some subsystems have there own CI
system which tests every patch. Does it build? Does it add new
warning, is sparse happy, are there more self test failures. What do
different AIs think of the patch etc. Cross subsystem patches don't
always hit the per subsystem CI tests.

So in general, it is better to submit per subsystem.

       Andrew