Less specification, more formal grammar. Foo-to-HTML converter is next to useless, especially compared to a well-formed parse tree. (In fact, there are already some PEG parsers for Markdown.)
It's not possible to write a formal (non-PEG) grammar for Markdown. Hence, a spec, which is the next best thing (a la HTML5). Of course, vfmd is not a md-to-html converter, it's an md-to-parse-tree converter + a parse-tree-to-html converter.
PEG is not a formal grammar? PEG being not a subset of context-free grammars? That's all new things to me.
And you actually have seen PEG parsers for Markdown? Like this one: https://github.com/jgm/peg-markdown/, with grammar written explicitly and lexer/parser generated?