I think his issue may be more that it's hard to be certain of this difference because it's relatively small and there are many other factors at play. He also doesn't say how many times he ran this test, but assuming he does sufficient test runs then I'd agree a 300-600ms difference is worth it.