Worst-case vs Average-case Design for Estimation from Fixed Pairwise Comparisons