It's hard (not impossible) to make it work in principle because a printing press doesn't need to infer meaning from the post script or whatever you provide it. A browser does need to, so you'd need to layer the extra semantics on top of your WYSIWYG editor.

Which means asking questions like "this is 36 pixels high: is it a heading,; or is this just a landing page with big text?" etc. So maybe I lack imagination, but I don't see it being that easy a problem to solve from a UI point of view.