People forget the weird workflow in any journal that does biomedical work (including most of the PLOS family): manuscripts are submitted as Word files, then converted to XML and reformatted by a semi-automated process. There's manual labor involved to format the references and cross-references, mark up tables and figures, and proofread the HTML, PDF, and XML versions. The XML is then submitted to PubMed Central for archival storage.
There's a lot of manual labor involved in the profess. You can't just hand-wave away and say "peer review is provided for free" and assume the rest of the process is free.
Proper formatting makes it possible to automatically extract bibliographic reference data, text-mine the contents for whatever interesting purposes you might think of, display it in apps and new formats, and store it for decades in a stable format.
Nice value added by the MSRI style right here.
There's a slow-moving Scholarly Markdown project, but I don't know what its status is. http://scholarlymarkdown.com/
I agree with the sentiment, though. If there were some markup language that supported what academics do with LaTeX and Word, and converted cleanly to JATS XML and to nice-looking PDFs, it would be enormously useful.
It seems much more likely that PLOS simply charges the "$1000 or so" because they can, and it's easy enough to piggyback on the traditional publisher narrative that they are actually providing this amount of added value.