Periodic molecular dynamics simulations of proteins may suffer from image interactions. Similarly, the hydrophobic effect required to keep a protein folded may not be enforced by small simulation cells. Accordingly, errors may arise both from the water concentration per se and the image interactions. Intrinsically disordered proteins are particularly sensitive, providing a worst-case estimate of the errors. Following this reasoning, we studied Aβ (Aβ), a disordered peptide central to Alzheimer's disease, by 100 different simulations with variable cell size from very large (20 Å) to very small (3 Å). Even for this very disordered peptide, most properties are not cell-size dependent, justifying the common use of modest-sized (10 Å) cells for simulating proteins. The radius of gyration, secondary structure, intrapeptide, and peptide-water hydrogen bonds are similar relative to standard deviations at any cell size. However, hydrophobic surface area increases significantly in small cells (confidence 95%, two-tailed t-test), as does the standard deviation in exposure and backbone conformations (>40% and >27%). Similar results were obtained for the force fields OPLS3e, Ambersb99-ILDN, and Charmm22. The similar prevalence of structures and α-β transitions in long and short simulations indicate small diffusion barriers, which we suggest is a defining hallmark of intrinsically disordered proteins. Whereas hydrophilic exposure dominates in large cells, hydrophobic exposure dominates in small cells, suggesting a weakening of the hydrophobic effect by image interactions and the few water layers available to keep the protein compact, with a critical limit of 2-3 water layers required to enforce the hydrophobic effect.