The velocity distribution of galaxies in clusters is not universal; rather, galaxies are segregated according to their spectral type and relative luminosity. We examine the velocity distributions of different populations of galaxies within 89 Sunyaev Zel'dovich (SZ) selected galaxy clusters spanning $ 0.28 < z < 1.08$. Our sample is primarily draw from the SPT-GMOS spectroscopic survey, supplemented by additional published spectroscopy, resulting in a final spectroscopic sample of 4148 galaxy spectra---2868 cluster members. The velocity dispersion of star-forming cluster galaxies is $17\pm4$% greater than that of passive cluster galaxies, and the velocity dispersion of bright ($m < m^{*}-0.5$) cluster galaxies is $11\pm4$% lower than the velocity dispersion of our total member population. We find good agreement with simulations regarding the shape of the relationship between the measured velocity dispersion and the fraction of passive vs. star-forming galaxies used to measure it, but we find a small offset between this relationship as measured in data and simulations in which suggests that our dispersions are systematically low by as much as 3\% relative to simulations. We argue that this offset could be interpreted as a measurement of the effective velocity bias that describes the ratio of our observed velocity dispersions and the intrinsic velocity dispersion of dark matter particles in a published simulation result. Measuring velocity bias in this way suggests that large spectroscopic surveys can improve dispersion-based mass-observable scaling relations for cosmology even in the face of velocity biases, by quantifying and ultimately calibrating them out.