They make pages.
It’s as simple as that, really. Academics like citing to pages, and HTML documents don’t give them that ease of reference.
As a bonus, being able to download an article as a PDF means you have access to the document exactly as it was intended to be published for a hard-copy audience. As long as articles are printed on paper, that’s useful for creating unique citation identifiers.
This could be changed by citing to paragraphs – as is the case in some legal scholarship – but authors and publishers would have to be willing to toss a lot of numbers into documents. And readers would have to be willing to ignore irritating clutter that breaks the flow of information.
For now, PDFs make a good transitional format because everyone can read them (as long as they aren’t password-protected). Too many changes at once, especially in areas that are only incidental to open access, just create barriers for those unwilling to leave paper and page citations behind.
The common sense solution, of course, is to have repositories which offer articles in a variety of formats, be they HTML, PDF or something else. Many are doing this already. As a result, readers have quick access to the text without firing up Acrobat Reader or some other viewer, while also providing an ‘official copy’ to be used for citations.