This whole "abstract" vs "physical" pixel thing (which CSS has also embraced) feels like a massive kludge to me. It's really weird to define the standard measurement in terms of "this is roughly 1/96 inch, potentially depending on how far from the user the screen is planned to be".
If that really is the most sensible measurement to use, then terminology clearer than "abstract pixel" should be used. Because people have an idea of what a pixel is, and it ain't that.
I like the idea of using something like CSS's 'em' as a base unit. That also ensures interfaces are likely to scale sanely if a user could benefit from an unusually large font size.