I don't know much about quantile estimation, so possibly dumb question: why, in the experiments plotted here, does every estimator on every tested distribution

I don't know much about quantile estimation, so possibly dumb question: why, in the experiments plotted here, does every estimator on every tested distribution underestimate the true median? I expected some errors, but not all systematically in the same direction.