The PR actually checks that the buffer is read-only (this was also a concern that Mark Shannon had).  And the Python buffer protocol says that you need to consistently hand out read-only buffers.  So while someone could create a buffer and mutate it outside of the buffer protocol it should be really read-only.  

As far as the ref counts, it's not just the ref counts for the code byte strings that are potentially problematic.  But it's the ref counts on all of the random other objects which the code objects are on the same page as, as well as other random read-write data that could be on those pages.  There's also an additional benefit in that code objects not loaded before forking can continue to share their memory as well.