It is a well-established fact that intranuclear-cascade models generally overestimate the cross sections for one-proton removal from heavy, stable nuclei by a high-energy proton beam, but they yield reasonable predictions for one-neutron removal from the same nuclei and for one-nucleon removal from light targets. We use simple shell-model calculations to investigate the reasons for this deficiency. We find that a refined description of the neutron skin and of the energy density in the nuclear surface is crucial for the aforementioned observables, and that neither ingredient is sufficient if taken separately. As a by-product, the predictions for removal of several nucleons are also improved by the refined treatment.