Below is an example that illustrates why we should use Kendall's tau instead of Pearson's r. The plot shows group A and B's quality assessments for server models of target T0370. I'm sure we all agree that the quality assessment of group B is much better than group A. However, Pearson's r for group A's assessment is slighter higher than for group B's assessment (r_A=0.90 vs r_B=0.89). If we use Kendall's tau for these two assessments, then we get t_A=0.22 vs t_B=0.67, which agrees with our intuition.