In the biomedical sciences (or any field that ends up on PubMed), articles have to be converted to JATS XML (http://jats.nlm.nih.gov/), a standard XML dialect for journal articles. It builds in citation metadata, cross-referencing, figure references, etc., and is supposed to be a stable archival format for long-term storage of articles. Individual publishers (PLOS, BMC, etc.) build their entire publication workflows around JATS, so articles can be typeset into PDF or rendered to HTML, or delivered to e-reader apps or whatever. Since it's semantic XML, you can do bibliography mining, automatic reference following, extraction of figures, or whatever you might want to make reading or text-mining easier.
But articles are often written in Word, so there's a tremendous amount of work going into manually or semi-manually converting every manuscript to semantic XML from the Word soup it arrives as. Same goes for LaTeX: a few journals just publish LaTeXed PDFs directly, but big publishers like Elsevier and Springer have semi-automated processes for converting LaTeX to in-house formats so they can provide HTML versions of pages.
So, short version: an editor supporting JATS XML can support all the features you need in a scientific document, and can dramatically simplify the publication workflow and hopefully save a bunch of money. And hopefully open-access journals pass that savings on to users.
For users, it could mean better e-reading apps (so you don't have to zoom in on tiny fonts in a PDF on your iPad), better support for cross-referencing and figures than Word has, automated formatting (journals style the XML, so you don't have to do margin and layout crap), and a simpler submission process.