In inverter-dominated microgrids with high renewable penetration, frequency and voltage regulation are strongly affected by load steps, source variability, and reduced inertia, which raises the question of how different control strategies trade off tracking accuracy, dynamic stress, power quality, and control effort under identical operating conditions. We conducted a systematic and reproducible comparison of six representative controllers, including active disturbance rejection control, sliding mode control, model predictive control, fuzzy proportional--integral--derivative control, and two lightweight learning-assisted approaches based on extreme learning machines and least-squares support vector machines. All controllers were evaluated on the same control-oriented microgrid model using stochastic renewable profiles, step load disturbances, measurement noise, and multiple Monte Carlo realizations. Performance was assessed using quantitative metrics covering frequency tracking, transient response, rate of change of frequency, voltage deviation, harmonic distortion, and control activity. The results show that fuzzy proportional--integral--derivative control achieved the best overall tracking performance with a root mean square error of 0.00753 Hz and an integral absolute error of 0.05586 Hz, while maintaining a moderate control effort of 0.200 p.u. Sliding mode control yielded the lowest voltage total harmonic distortion at 3.30\%, whereas model predictive control produced the smoothest control signal with a mean control effort of 0.017 p.u., but it suffers from significantly larger tracking errors, with a root mean square error of 0.07488 Hz. These results demonstrate that controller performance is strongly metric-dependent and that no single strategy is universally optimal for all operational objectives.