I can imagine this as a viable option e.g. for loading external code. You don't write one monolithic program that is >64K. You write several smaller ones. You don't jump between pages like crazy - changing a page should be a thought out and careful operation. Either only part of the memory (say, upper 32K) is swapped, or there exist large overlap areas of the program. This way switching the "memory banks" should look transparent from the program point of view. You disable interrupts, perform MOV P3,#2 and switch the first external memory chip off, the other on. Several NOPs to let things stabilize, reenable interrupts and then continue execution from memory which has changed "under your feet". Completely new program, maybe the same, maybe different interrupt routines. Compiled with the same compiler as usually, just containing pieces of unusual code.