In TM systems, the phenomenon of performance loss due to high concurrency can be partially prevented by bounding the number of threads to the number of available CPU-cores [3]. However, such as shown in Figure 1, the performance degradation may arise even when there are less concurrent threads than CPU-cores.