Our stop()/start() API isn't the greatest. Yehuda had some ideas for making that more convenient, outlined here: https://gist.github.com/2138c7bf20da83e823e8

Worth noting as well: nodeunit passes in a test object to each test, where the test has to call test.done() in the end. Each test gets its own instance, so there's no risk of one async test bleeding into another, like it currently can happen in QUnit. We should look into providing something similar for QUnit.