The performance evaluation of motion planning algorithms for agricultural robotic manipulators is commonly performed via benchmarking platforms. However, creating a realistic benchmarking scene that constrains the motion planning algorithms with the characteristic of a real-work environment has always been a challenge worthy of research. In this paper, we present a lab-setup benchmarking platform to evaluate Open Motion Planning Library (OMPL) motion planners for the application of a robotic harvester of a palm-like tree using a real-time 3D occupancy grid map. First, three motion problems were defined with different levels of complexity based on a real oil palm fruit harvesting task. To achieve reliable outcomes, the benchmarking scene was modeled by converting point cloud data from a stereo-depth sensor into a 3D occupancy grid map using the Octomap algorithm. Then the benchmarking was performed, all within a real-time process. According to the results, a fair performance evaluation was achieved by modeling a realistic benchmarking scene, which can help in choosing a highperforming algorithm and efficiently conducting such harvesting tasks in real practice.