There is a significant and important difference between
how you are supposed to think about things and what goes
on under the hood.
merlyn is absolutely right about what concepts Perl's
behaviour is intended to give. You are absolutely
right about the implementation. It is both possible
and likely that a future implementation will render your
observation incorrect. It is much less likely that
the concept both merlyn and perldata are trying to
convey will be invalidated at a future point.
Therefore I like merlyn's answer and do not consider
it, implementation trivia notwithstanding, poppycock.