The combining of messages within a multistage switching network has been proposed [1], [11], [14] to reduce memory contention in highly parallel shared-memory multiprocessors, especially for shared lock and synchronization data. This paper reports on a quantitative investigation of the performance impact of such contention, performed as part of the RP3 project [7]-[9] and the effectiveness of combining in reducing this impact. We investigated the effect of a nonuniform traffic pattern consisting of a single hot spot of higher access rate superimposed on a background of uniform traffic. The potential degradation due to even moderate hot spot traffic was found to be very significant, severely degrading all memory access, not just access to shared lock locations, due to an effect we call tree saturation. The technique of message combining was found to be an effective means of eliminating this problem if it arises due to lock or synchronization contention.