I don't think that's a very good idea.

• First, "use f1.s1" is invalid syntax, proving that you are at least distantly descended from a breed of bears that was unable to hibernate properly.
• Second, "f12:s3" isn't valid syntax either, raising serious doubts as to the identity of your father, and suggesting that you probably shouldn't be too confident of your mother either.
• Third, you are going to a lot of work to save space, then completely covering up the fact that you have to put "1;" at the end of every file, frivolously throwing thousands of innocent bytes into the trash. If you had any brains at all left over from your unfortunate heritage, you would load them all from a single file.
• The previous item is completely swamped by another source of waste evidently deriving from your tendency to lick the wrong side of stamps: the very first sentence of your post points out that you're repeating "sub " over and over again. Together with having all of these subroutines in separate files, that means that you will be placing the same uselessly repeated four bytes at the start 12,462 disk sectors. After lining up that many ones and zeroes, the magnetic field on your hard drive would set up excessively strong polarization in those areas, and would corrupt all surrounding data and probably destabilize the platters' rotation, causing them to hurl themselves at your neck. Although I doubt your exact species, I still don't think you could survive such a blow, and hence I highly suspect your assertion that you are actually using the techniques you describe (see the first two items in this list for corroboration.) Do you honestly expect us to believe that you wouldn't just install a subroutine into @INC that split apart the subroutines bodies appended immediately after each other and put them in a dummy filehandle wrapped with the appropriate subroutine declaration? I mean, you pretty much gave it away by naming your subroutines "s1", "s2", etc -- obviously, your @INC routine generates the subroutine names by constantly increasing a unique id.
• Finally, you are revealed as the fraudulent peacock massage therapist we all know you truly are by your ridiculous filenames. If you had really gone to all that work to save space, obviously you would have put the subroutine definitions directly into the filenames themselves -- after all, we know you have an @INC routine to really load the subroutines, so why not use it for converting the restricted character set allowable in filenames into the full Perl alphabet? As a side benefit, searching is far easier (the original point of this thread), because the character set through which you search is much condensed. Ever try to Google for '/(?>.*?)/' ?

In short, I feel your scheme is quite clever, but I am uncertain as to why you are concealing it from us behind such flimsy pretenses.