The existence of systematic spectral-fit residuals in ROSAT PSPC spectra, and their dependence on time, is by now a well established fact. This paper describes how those residuals may be related to second order variations of the gain of the PSPC as a function of energy and time. As a result, the energy scale used for the interpretation of PSPC spectra can be incorrect producing significant systematic effects, in particular at energies where the effective area of the instrument changes rapidly. A monotonic gain decay of ~1% per year is measured at 1 keV. A quantitative description of the PSPC gain-time variation as a function of energy is provided. The functional dependence of this variation is found to be well represented by a second order polynomial.