vice-versa sub project: css2xpath

Update

I completely re-wrote the test suite. You can check a live example here where results are compared with Sizzle selector engine against the common W3C page.

---

After this Ajaxian post and some opinion exchange in its comments, I decided to extract the css2xpath function I used inside experiments in vice-versa project, creating a separated project specific for this purpose that some how could be interesting or useful.
Bugs, problems, questions, and helps are more than welcome, so far I focused my attention to SlickSpeed test selectors and nothing else but at least for that it seems to be stable and reliable.

The Fastest Selector Engine For FireFox 3.0 ?

I am putting some effort to make vice-versa project a good, production ready, alternative against common libraries, and since these days I am working hard with XML, XSL(T), and XPath I decided to try an experiment implementing CSS to XPath translator for FireFox 3.0



With version 3.1 we will have querySelector and querySelectorAll but version 3.0 is still the most used one, FireFox speaking!

The nightly build of vice-versa introduces a new file, mainly used for personal experiments via vice-versa library and the first experiment is an improved document.query for those browsers with document.evaluate support without document.querySelectorAll.
Unfortunately Internet Explorer < 8 does not support XPath queries over (x)HTML but at least for FireFox 3.0 I have obtained best overall performances via SlickSpeed Selector Speed Test comparing latest version of each library such Dojo, DOMAssistant, Sizzle (congrats for the new site!), Sly, and finally vice-versa project.
Here the summary, from faster to slower:

vice-versa    760ms
Dojo 1.3.1    868ms (2 tests failed)
DOMAssistant  880ms (good stuff Robert!)
Sly           890ms
Sizzle        914ms

Now, the good part is that nowadays every selector engine seems to perform truly fast and in whatever browser you like, but the "even better" part is that the experiments file in vice-versa is 1Kb (minified and gzipped) and with just "that Kb" it is possible to use a specific and fast selector engine for FireFox 3.0.

Finally, the experimental CSS to XPath translator cannot support pseudo searches like :hover, :active, or similar, due to the XPath nature, a query language mainly designed for XML where CSS, unless we do not transform it via XSLT, has a complete different meaning from the one used in (x)HTML.

I wonder now who will be able to create better overall performances for FireFox 3.0 ( changing my translator implementation, for example ;) )