This study presents a comparative evaluation of four metaheuristic algorithms: Cuckoo Search (CS), Dwarf Mongoose Optimization (DMO), Butterfly Optimization Algorithm (BOA), and the proposed Butterfly Optimisation Algorithm 2 (BOA2) for multi-objective routing optimisation in Mobile Ad Hoc Networks (MANETs). Simulations were conducted in MATLAB under varying node densities (20, 50, 100, 200, 500, and 1000 nodes). Performance was evaluated using Packet Delivery Ratio (PDR), throughput, end-to-end delay, and standard deviation to assess robustness and stability. Results show that network density significantly influences algorithm performance. At 20 nodes, BOA and BOA2 achieved the highest PDR (37.33% and 37.38%) and throughput (30.66% and 30.70%), along with low delay values, indicating efficient routing in less congested environments. As node density increased, performance differences became more pronounced. CS experienced substantial degradation in both PDR and throughput, demonstrating limited scalability. DMO showed moderate improvements over CS but did not maintain consistent superiority at higher densities. Although BOA remained competitive at 50 nodes, its performance declined at 100 nodes. BOA2 consistently achieved the best overall performance at higher densities, recording the highest PDR (18.65%) and throughput (15.32) at 100 nodes while maintaining competitive delay. Performance recovery was observed at higher densities (200 and 500 nodes), attributed to improved connectivity and route diversity, and BOA2 consistently achieved the highest PDR and throughput. Under ultra-dense conditions (1000 nodes), overall performance declined due to congestion and routing overhead; however, BOA2 maintained comparatively stable performance across all metrics. These results confirm that adaptive enhancements in BOA2 improve exploration–exploitation balance, scalability, and routing reliability, making it the most effective algorithm for dense MANET environments.