Physics-compliant channel models for radio environments parametrized by reconfigurable intelligent surfaces (RISs) require the inversion of an "interaction matrix" to capture the mutual coupling between wireless entities (transmitters, receivers, RIS, environmental scattering objects) due to proximity and reverberation. The computational cost of this matrix inversion is typically dictated by the environmental scattering objects in non-trivial radio environments, and scales unfavorably with the latter's complexity. In addition, many problems of interest in wireless communications (RIS optimization, fast fading, object or user-equipment localization, etc.) require the computation of multiple channel realizations. To overcome the potentially prohibitive computational cost of using physics-compliant channel models, we i) introduce an isospectral reduction of the interaction matrix from the canonical basis to an equivalent reduced basis of primary wireless entities (antennas and RIS), and ii) leverage the fact that interaction matrices for different channel realizations only differ regarding RIS configurations and/or some wireless entities' locations.