If I remember correctly when you use a function from a module the entire
module is loaded into memory? Thus, if your module is smaller it may run
faster.

For readability it's a good idea to put code into separate modules
anyway(eg. all functions that alter an employee record).

Theoretically splitting code up into smaller sections should put less of a
strain on the system.  If you have 80 subs and functions in one module, then
if you call one, the whole lot gets loaded, whereas perhaps you only needed
a few.
Having said that, I have never really noticed a difference and my modules
are built up by logically organizing my code into blocks.  In other words,
readability and logic is my priority.  However, you are now in a position to
do some empirical tests.  Take 2 versions of the database and make some
measurements - then *you* can let *us* know, what practical difference you
measured.

I think it's probably a result of discarding old compiled code and
getting freshly compiled code to replace it.

If you coded with COMPILE ON DEMAND off all the time, you'd get the
same benefit.