He didn't show that full scans are better than indexes, just that they aren't obviously worse when the data is small.