The topic of traffic signal control (TSC) for urban intersection networks has been a key research area for many years, but comprehensive comparisons among state-of-the-art methodologies remain limited. This calls for a more exhaustive approach to benchmarking TSC techniques. This study builds upon previous TSC benchmarking work by evaluating the performance of both reinforcement learning (RL) and model-based TSC algorithms across a spectrum of demand levels. We adapt new metrics for in-depth simulation analysis and ground our investigation in a real-world urban arterial scenario. Our results unveil a noteworthy observation: the performance of tested algorithms is inconsistent when evaluated under diverse demand levels and across different metrics. This highlights the importance of implementing multidimensional evaluations in future TSC studies for a more nuanced understanding of their performance.