Estimating the expected size of the largest earthquake on a given fault is complicated by dynamic rupture interactions in addition to geometric and stress heterogeneity. However, a statistical assessment of the potential of seismic events to grow to larger sizes may be possible based on variations in magnitude distributions. Such variations can be described by the b-value, which quantifies the proportion of small- to large-magnitude events. The values of b vary significantly if stress changes are large, but additional factors such as geometric heterogeneity may affect the growth potential of seismic ruptures. Here, we examine the influence of fault roughness on b-values, focal mechanisms, and spatial localization of laboratory acoustic emission (AE) events during stick-slip experiments. We create three types of roughness on Westerly granite surfaces and study AE event statistics during triaxial loading of the lab faults. Because both roughness and stress variations are expected to influence b, we isolate roughness contributions by analyzing AEs at elevated stresses close to stick-slip failure. Our results suggest three characteristics of seismicity on increasingly rough faults: (1) seismicity becomes spatially more distributed, (2) b-values increase, and (3) focal mechanisms become more heterogeneous, likely caused by underlying stress field heterogeneity within the fault zones. Localized deformation on smooth faults, on the other hand, promotes larger rupture sizes within the associated homogeneous stress field, which is aligned with the macroscopic stress orientation. The statistics of earthquake magnitude distributions may help quantify these fault states and expected rupture sizes in nature.