What’s he’s saying (behind too many opinions) is that actual HCI studies collected in something resembling a scientific manner show very clearly that skeuomorphic work better, for many clearly defined metrics of better.