I don't know the guts of the MVS or VM implementations of Perl. From my memory of working on these systems, I believe a Perl program would be run as a user process. As such, the user's Perl program would be data to the Perl interpreter, and the memory used to store the variables would be distinct from the memory used to store the instructions generated from the user's program. Both would be distinct from the perl interpreter, itself. Buffer overflow would almost certainly cause the user's process to be immediately aborted with the dreaded 0C4 error, which was, if I remember, not catchable: control would absolutely not be returned to the application.

---

added in update The VM to which I'm referring is IBM's mainframe OS.