He didn't give any example code or raw data (which I'd love to see), but my guess is that the 21st land resulted in playing more of his spells on curve. The 20 land version likely had a statistically higher chance to get stuck with cards in hand but no mana to cast them than the 21 land version.