In this chapter, we evaluate the process for optimising partitioned temporal joins - as proposed in chapter 6 and elaborated in chapters 7 to 9. The experiments will focus on the various features that have been discussed so far and will follow the same path as the preliminary evaluation in section 8.5 which assumed uniform test data. Here, we will use real temporal data that was extracted from existing temporal relations; section 10.1 describes these data sets in more detail. In section 10.2 the performances of the various strategies of chapter 9 are compared in order to identify the most promising ones on which we can concentrate in the remaining experiments. This reduces the complexity of the following experiments significantly and makes the results easier to visualise and to interpret. In section 10.3, we look at the problem whether it is better to partition a join into many small or into a few but slightly bigger join computations. In other words, we vary the parameter m . While m is an input parameter for the uniform partitioning strategies it is an output parameter for the underflow and minimum-overlaps strategies. There, it is imposed by parameters X or XR , XQ which set maximum sizes for the various fragments or subfragments. In section 10.4 we therefore try to find a rule that allows us to determine the best-performing value for these parameters. In section 10.5, we look at the influence of the average interval length on the performances and whether certain values of favour certain partitioning strategies. In section 10.6, the sizes of the participating relations are varied. In section 10.7, we return to the question of the best-performing mixture / of SMP-nodes and processors per SMP-node. We have already looked at this problem when conducting the experiments for uniform data. Here, we use skewed and somehow more realistic data. In sections 10.8 and 10.9, we look at the influences of condensation and black-out preprocessing on the performances. Finally, the main results are summarised in section 10.10.