Molecular dynamics simulations have become crucial in various important research areas, such as thermodynamics, structural mechanics, and medicine. Since the effectiveness of such simulations in research is heavily dependent on their accuracy and fast availability, there is a continuous effort to optimize the algorithms involved in their making. Multi-site simulations form a subdiscipline within the molecular dynamics landscape, where every molecule gets treated as a rigid body of points. This thesis investigates the efficiency of different molecule representations and data structures that can be used to represent the molecules in such simulations. We show that the molecule representation chosen can significantly affect the runtimes of the molecular dynamics algorithms acting on it. The relation between molecule representation and molecular dynamics algorithm is captured by testing different combinations of the two in various simulation scenarios. We show that the introduction of new variables to reduce the computational workload of the torque evaluation can achieve a speedup of up to 1.4 depending on the scenario, whereas the explicit storage of site position actually decreases efficiency. Additionally, we discuss another approach, where every site gets treated as an independent entity during the force calculation process. We highlight that the efficiency of this representation strongly deviates in both directions from a more conventional approach depending on the data-layout and neighbor identification algorithm. We discuss this relation in greater detail.