I don't think it was 'rules' or a decision as such, but more a case of early limitations to possibilities of implementing one technology 'inside' another.
It was likely to have been technical neccesity that dictated that plug-in content would be windowed into a different channel to that of native content.
Possibly something to do with the fact(?) that the underlying technology in html rendering hasn't really changed since day one and the way that plug-in content is by nature 'added on' later to the browser's native capabilities.
Also, it's worth remembering that dhtml and floating layers (z-indexed) are a relatively recent addition to the tricks a browser can do. As such, I don't imagine they were ever part of the original blueprint and therefore there was no reason to believe that plug-in content would be problematic if kept at the top of the stack (so to speak).
That's my semi-educated guess, anyway
I can only guess that the fact that MSIEWin handles this 'correctly' have managed to get ahead of the pack for once.
(Now if they could only pull their collective thumb from out of their arse regarding fundamentals...)