Optimal operation of chemical plants is usually accomplished by first finding the optimal steady state using the present (nominal) set of disturbances. This is usually implemented by sending constant setpoints for selected variables to the control system. The setpoints are then the nominally optimal values. However, because of disturbances this may result in feasibility problems, which we here try to avoid by adjusting the setpoints ("backoff"). First, we need to avoid infeasibility in the active constraints ("constraint backoff"). Second, we need to adjust the setpoints of the unconstrained controlled variables. This may be done by offline computation of robust setpoints ("optimal backoff") or by online feasibility correction ("flexible backoff"). As a case study, we consider a reactor-separator-recycle process. For this process, the control structures based on Luyben's rule ("fix a flow in every recycle loop") are infeasible if we use the nominal setpoints but are feasible with reasonable loss if we use robust setpoints.