Uncrewed aerial vehicle (UAV) path planning in complex environments requires more waypoints to generate high-quality flight paths. However, increasing waypoint density significantly raises computational complexity and risks slow convergence or entrapment in local optima. Although Artificial Bee Colony (ABC) algorithms are commonly employed for UAV path planning, they often struggle to achieve efficient exploration and exploitation, especially in scenarios with frequent terrain height variations. To address these limitations, we have proposed a Multi-Dimensional Perturbation Artificial Bee Colony (MDP-ABC) algorithm. The multi-dimensional perturbation strategy is introduced in the employed bee phase to balance exploration and exploitation through preferential and random selection. In the onlooker bee phase, the curvature-guided elite neighbor search strategy is used to prioritizes high-curvature waypoints, enhancing optimization efficiency in complex terrain. Furthermore, path costs are independently modeled in the horizontal and vertical directions. The MDP-ABC algorithm has been validated in three scenarios with different complexity, and compared with other ten algorithms. Simulation results demonstrate that MDP-ABC significantly enhances convergence speed, solution quality, and robustness, achieving an average performance improvement of 71.9% over ABCiff and 93.2% over AFT in the complex scenarios. These results confirm the MDP-ABC algorithm with the great effectiveness for solving the UAV 3D path planning in the complex environments.