We demonstrate the importance of properly accounting for London dispersion and basis-set-superposition error (BSSE) in quantum-chemical optimizations of protein structures, factors that are often still neglected in contemporary applications. We optimize a portion of an ensemble of conformationally flexible lysozyme structures obtained from highly accurate X-ray crystallography data that serve as a reliable benchmark. We not only analyze root-mean-square deviations from the experimental Cartesian coordinates, but also, for the first time, demonstrate how London dispersion and BSSE influence crystallographic R factors. Our conclusions parallel recent recommendations for the optimization of small gas-phase peptide structures made by some of the present authors: Hartree-Fock theory extended with Grimme's recent dispersion and BSSE corrections (HF-D3-gCP) is superior to popular density functional theory (DFT) approaches. Not only are statistical errors on average lower with HF-D3-gCP, but also the convergence behavior is much better. In particular, we show that the BP86/6-31G* approach should not be relied upon as a black-box method, despite its widespread use, as its success is based on an unpredictable cancellation of errors. Using HF-D3-gCP is technically straightforward, and we therefore encourage users of quantum-chemical methods to adopt this approach in future applications.