From my reading of the responses to my question I see two ways that people are thinking of proceeding, which I think might complement each other well.
First, there's the approach of fixing the current pages by moving the style information out of the Python code and into a CSS file. Second, there's the more extensive work to switch to a proper templating system like Genshi. I think in the long run, getting all the XHTML into external files is where the project should go. That will give users the most freedom to redesign how they see fit (good design of the default templates would still have that be styled using CSS), but it looks like there are a lot of work to make good progress in that area. I'd like to suggest a mixed approach. If we started to working fixing the 2.1 branch to have better output (even if it's never properly released), that would allow some people to do useful work while Barry firms up the trunk and we make a final choice of a templating engine. Once a template engine is inserted, that improved output could serve as the basis for the first templates for the system. Also, while we're finding a changing the HTML in the current python code, we could mark that code to be easy to find later.
We also need to attend to the "creaky decade-old look", but I would suggest that really be taken on as part of the work to move to a real template system, not as part of picking off the low hanging fruit.
Aaron