They don't cover the exact same date range (20y has 10 more years of 20y returns than the 30y), plus it's a median so it could be randomness that skewed it towards a higher return vs. the 30y. (The 10.21% return happened 1 year before the first 30y return is recorded.)
Interesting that if you take the mean you get: 10y, 20y, 30y :: 8.87%, 9.67%, 10.23%
Increasing mean returns as you increase holding period, and the mean 30y is almost exactly the median 20y.