[thepsi]

Solving it perfect is non-trivial (I've known entire PhDs to be spent working on a small subset of the problem). There are a number of products/projects that solve it to some extent (techniques include absolute positioning & making sweeping assumptions about what constitutes a paragraph) - would this be enough for you to consider paying for, given that their assumptions/workarounds might produce HTML files that aren't quite to your 'taste'?