This study presents a controlled system-level benchmark of a Building Information Modeling (BIM)-derived indoor IEEE 802.11ax deployment under variable user density and access-point (AP) association policy. Four fixed APs, inherited from a preceding radio-frequency (RF) coverage optimization, were evaluated with 10, 30, 50, and 70 static stations under received signal strength indicator (RSSI), load-aware, and hybrid association. The primary benchmark used fixed modulation and coding scheme (MCS) 4 with link adaptation disabled, 2.4 GHz/20 MHz operation, a fixed 1-6-11-1 channel plan, and downlink full-buffer traffic. Since BIM wall mapping was partial, an empirical fallback of 50 dB/m was applied to unmapped segments, and this assumption was tested in a low-loss foam sensitivity analysis. Across 120 paired-topology runs, association policy did not significantly change aggregate throughput (p=0.111), but it strongly affected Jain fairness, station-count load imbalance, and serving RSSI. At 70 stations, all policies entered a functional congestion collapse, with approximately 62% mean packet loss and 15% zero-throughput users. Supplementary analyses indicated that station-count balance alone was not sufficient to infer per-AP delivered-throughput balance under saturated traffic, while a three-seed bidirectional bursty test preserved the same qualitative policy ordering. The results support load-aware association as a resource-distribution mechanism, but not as a remedy for configuration-specific saturation or RF-coverage limitations.