Challenge Accepted.

gistfile1.js

getEl = (function (doc) {

    var getElementsByClassName = (function () {

        return doc.getElementsByClassName ?
        function (selector) {
            return doc.getElementsByClassName(selector.split('.').pop());
        } : function (selector) {

            var parts = selector.split("."),
                tags = document.getElementsByTagName(parts[0] || "*"),
                matches = [],
                classes, i, iLength, c, cLength;


            for (i = 0, iLength = tags.length; i < iLength; i++) {
                classes = ("" + tags[i].className).split(" ");
                for (c = 0, cLength = classes.length; c < cLength; c++) {
                    if (classes[c] == parts[1]) {
                        matches.push(tags[i]);
                        break;
                    }
                }
            }
            return matches;
        }

    })();

    return doc.querySelectorAll ?
    function (selector) {
        return doc.querySelectorAll(selector);
    } : function (selector) {
        var id;
        if ((id = selector.split("#")).length > 1) {
            return document.getElementById(id.pop());
        } else {
            return getElementsByClassName(selector);
        }
    }

})(document);

Ah, nice, but I don't know if getElementsByClassName works for this use case.. ideally they'd have a tag name, too. I guess it's nice to offer the flexibility though. HMMMMM.

Sounds like you need sizzle?

Don't think I need the whole thing.. Just have to allow #id or tag.class - should always be one of those. I guess I could take part of it.

ooh, works with tag name http://jsfiddle.net/danheberden/rEnjn/ - actually, nvm :( fails on ie6,7

Yeah, works great when it uses queryselectorall. :D

NVM

Ha, I just removed that, cause I guess it's only faster in webkit.. Oops. But I forgot to add the pipe in the jsperf that the original gist had to prevent that problem.