Your work is interesting and sure looks like good ruby to me.

I had to write a similar parser for Zena (to parse textile additions
and the zafu templates) and I thought that doing so many regex
evaluations on the full text (can be long) was too slow (please tell
if this is wrong). I thus chose to use regex anchored left /\A.../ and
eat through the text only once. This has the other advantage that you
enter different modes (tag parameters, comments, raw data, etc) along
the way. It makes it very easy to parse sub languages from within
these modes.

The parser is a two step operation: 1. parse, 2. render. This might be
overkill for the kind of transformations you need but it is very
interesting because the parsed elements can use some knowledge from
the context when they are rendered.

I intend to do a textile -> Latex transformation so the users can
write zafu templates to generate PDF.

The current implementation of the parser is not as clean as yours but
works. You can have a look at the parser here :
http://dev.zenadmin.org/browser/trunk/lib/parser

Let me know what you think.

Gaspard