Unfortunately, if you saw only a 100% improvement going from $$ to document.getElementsByClassName, I’m not sure the further optimizations would be significant enough. In browsers that support XPath, document.getElementsByClassName is now just as fast as (if not faster than) any other DOM query, but IE is not one of those browsers. Until IE supports DOM Level 3 XPath, there won’t be a no-drawbacks way to query the DOM in non-”standard” ways.

In the meantime, you can improve the performance of getElementsByClassName considerably if you restrict the query to a certain subset of the page: $(’container’).getElementsByClassName(’foo’) [or document.getElementsByClassName('foo', $('container'))]. This will cut down on the number of elements that must be searched through.

I’ve subsequently tried it, but it turns out to be not much faster on IE (perhaps a 2x speedup, whereas the server-generated JS gives something like a 20x speedup).

Maybe XPath gives a better speedup on IE.

I’m tempted to create a benchmarking exploration tool on ajaxify – a massive dom and a JS sketchpad to benchmark queries.

Please look into document.getElementsByClassName. It’s orders of magnitude faster, especially with the recent performance optimizations (including querying by XPath if it’s available). $$(”.foo”) can be replaced with document.getElementsByClassName(”foo”) in every instance. I guarantee that’ll fix your problems without forcing you to give up querying by class name altogether.

That said, making $$ less costly is one of the major points of focus for Prototype’s 1.5.0 release.

I don’t know if you have seen the following Microsoft posts, but I have found them really useful for making JS faster (particularly in IE)!

http://blogs.msdn.com/ie/archive/2006/08/28/728654.aspx

Ps: Currently reading your book Ajax book, it’s great!