Why not to go the other way round in this case? Typeset your thesis in semantically correct, searchable, beautiful HTML5, and then just add some specific CSS to get the hard print copy?
Unfortunately, browsers are not up to the job so far. For example, page-break-inside:avoid is only supported by Opera[0]. Maybe Prince [1], which is not for free, though.
I don't argue that TeX is good for typesetting thesis. I doubt that sphinx, even with some plug-ins is good for that. Also don't forget that reStructuredText has its own problems (not found in HTML and LaTeX). Can I have some bold italic or italic monospaced phrase example in reStructuredText please? No, really?
