S-AI-CITY is a first-generation hormonal orchestration framework for urban intelligence in which heterogeneous urban observations are transformed by six Gland Agents into a shared regulatory state composed of Mobilin, Energexin, Environin, Resiliencin, Normin, and Servicin. Article I introduced the architecture, while Article II formalized its multi-layer urban graph, reaction-diffusion hormonal dynamics, 24 specialized agents, sparse constrained orchestration, hierarchical Zone → District → City organization, and symbolic Urban Engram memory. The present Article III defines the empirical evaluation stage of the trilogy. The study is organized around the SAI-UT+ CITY testbench and twelve parameterized urban scenario families spanning nominal operation, sector-specific stress, multi-domain disturbances, cascading failures, degraded communication, missing data, recovery, and recurrent events. The evaluation compares S-AI-CITY with a ladder of non-hormonal, rule-based, learned, model-based, random-budget, and offline-oracle references under matched quality-of-service and safety conditions. Decision quality, parsimony, operational cost, hormonal reactivity, cross-domain coordination, resilience, memory acceleration, explainability, constraint satisfaction, robustness, and scalability are evaluated jointly. The experimental design is explicitly falsifiable: