I definitely vote for the PDF generation through another means option. I've done autogeneration of reports based on data, some pulled from flat files, some pulled from database backends, and it always seems easy to go through some language that affords a certain level of redundant markup. As such, the XML, HTML options are quite helpful, for markup, but formatting is a different issue.

Thus, I have to say I quite like using TeX for these types of things, because it gives you markup when you want it, it's language based so it's easy to generate from within scripts, and it will give consistent formatting if that is necessary (generating froms from DBs, etc.). It might take a little bit more work, since you need to be familiar with TeX, and it might take a few more CPU cycles, but it pays off by offering quite a bit more control over the layout.