I write most SRFIs in Markdown and then convert to HTML, but in other cases I take existing SRFIs such as SRFI 1 and heavily modify them. I wrote TagSoup, a Java program that converts arbitrary HTML to well-formed XML, so converting to XHTML would be easy, and I could do that conversion in a big batch.
Overall though I would prefer minimal modification with just a few classes, enough to make indexing straightforward, rather than a comprehensive reformatting effort. I could easily add class attributes by hand after HTML conversion. In any case I only convert once and then abandon the Markdown.