Background A critical issue in multicenter randomized trials focusing on surgical techniques is quality control, as the quality of the surgery usually varies widely if the procedure employed is complicated. Few studies have evaluated interinstitutional variation in randomized trials in order to check not only the generalizability of the results but also the reliability of the study group itself. Methods Two randomized phase III trials (JCOG9501 and JCOG9502) were conducted that compared standard and experimental surgery for gastric and esophagogastric junction adenocarcinomas. Mixed effects models were used to examine short-and long-term outcome data for 521 patients from 23 hospitals in JCOG9501 and 157 patients from 21 hospitals in JCOG9502. Results In both trials, some variation was observed in the number of dissected lymph nodes, the operative time, and the volume of blood lost. Estimated 5-year overall survival after standard surgery differed among hospitals (JCOG9501, 58.0-75.1 %; JCOG9502, 49.1-58.7 %), while there was little variation in the hazard ratio for overall survival (OS) for experimental versus standard surgery (JCOG9501, 1.05-1.48; JCOG9502, 1.44-1.48). Higher hospital gastrectomy volume was significantly correlated with a lower proportion of postoperative complications in JCOG9501 (q = -0.524, P = 0.010) and reduced blood loss in JCOG9502 (q = -0.442, P = 0.045). OS was not correlated with hospital or surgeon volume. Conclusions There was some degree of interinstitutional variation in outcomes after standard surgery, but there was little variation in the hazard ratio for OS for experimental surgery, indicating that the final conclusions of the two randomized phase III trials can be generalized to their respective target populations.