I can't tell if they're comparing apples to oranges here. For example, are the stats adjusted to account for the fact that self driving cars are only driving on roads that are not icy, whereas we human drivers drive in all kinds of conditions?
"Benchmarks were aligned to the same vehicle types, road types, and locations as where the Waymo Driver operated."
So yes, it's apples to apples... I'm sure there's some inherent variation, so maybe it's fuji to granny smith apples, but it seems there's nothing wildly off in their methodology.