In an important class of problems, the subsample bootstrap performs more poorly than asymptotic methods, even if the subsample size is chosen optimally. A technique related to Richardson extrapolation ...