The capacity of proteins to adapt their structure in response to various perturbations including covalent modifications, and interactions with ligands and other proteins plays a key role in biological processes. Here, we explore the ability of molecular dynamics (MD), replica exchange molecular dynamics (REMD), and a library of structures of crystal-ligand complexes, to sample the protein conformational landscape and especially the accessible ligand binding site geometry. The extent of conformational space sampled is measured by the diversity of the shapes of the ligand binding sites. Since our focus here is the effect of this plasticity on the ability to identify active compounds through virtual screening, we use the structures generated by these techniques to generate a small ensemble for further docking studies, using binding site shape hierarchical clustering to determine four structures for each ensemble. These are then assessed for their capacity to optimize enrichment and diversity in docking. We test these protocols on three different receptors: androgen receptor (AR), HIV protease, and CDK2. We show that REMD enhances structural sampling slightly as compared both to MD, and the distortions induced by ligand binding as reflected in the crystal structures. The improved sampling of the simulation methods does not translate directly into improved docking performance, however. The ensemble approach did improve enrichment and diversity, and the ensemble derived from the crystal structures performed somewhat better than those derived from the simulations.