The jsperf is broken because you accumulate event handlers across test boundaries. In other wrods, whatever test is run first will be the "fastest".
And the whole premise of the test is ridiculous, there is no difference between a function that has a name and a function that doesn't have a name if everything else is equal. You will only see a difference when you are setting up jsperf incorrectly. When you constantly get equal results for them then you know that you set it up correctly - but you would know this already from common sense :)