Deep heating of pelvic tumours with electromagnetic phased arrays has recently been reported to improve local tumour control when combined with radiotherapy in a randomized clinical trial despite the fact that rather modest elevations in tumour temperatures were achieved. It is reasonable to surmise that improvements in temperature elevation could lead to even better tumour response rates, motivating studies which attempt to explore the parameter space associated with heating rate delivery in the pelvis. Computational models which are based on detailed three-dimensional patient anatomy are readily available and lend themselves to this type of investigation. In this paper, volume average SAR is optimized in a predefined target volume subject to a maximum allowable volume average SAR outside this zone. Variables under study include the position of the target zone, the number and distribution of radiators and the applicator operating frequency. The results show a clear preference for increasing frequency beyond 100 MHz, which is typically applied clinically, especially as the number of antennae increases. Increasing both the number of antennae per circumferential distance around the patient, as well as the number of independently functioning antenna bands along the patient length, is important in this regard, although improvements were found to be more significant with increasing circumferential antenna density. However, there is considerable site specific variation and cases occur where lower numbers of antennae spread out over multiple longitudinal bands are more advantageous. The results presented here have been normalized relative to an optimized set of antenna array amplitudes and phases operating at 100 MHz which is a common clinical configuration. The intent is to provide some indications of avenues for improving the heating rate distributions achievable with current technology.