We extend our previous analysis which used generalized luminosity functions (GLFs) to predict the number of quasars and galaxies in low-radio-frequency-selected samples as a function of redshift, radio luminosity, narrow-emission-line luminosity and type of unified scheme. Our extended analysis incorporates the observed submillimetre (850 micron) flux densities of radio sources, employs a new method which allows us to deal with non detections, and focuses on the high-luminosity population. First, we conclude that the submillimetre luminosity L_{850} of low-frequency-selected radio sources is correlated with the bolometric luminosity L_{Bol} of their quasar nuclei via an approximate scaling relation L_{850} \propto L_{Bol}^{0.7 \pm 0.2}. Second, we conclude that there is quantitative evidence for a receding-torus-like physical process for the high-luminosity population within a two-population unified scheme for radio sources; this evidence comes from the fact that radio quasars are brighter in both narrow emission lines and submillimetre luminosity than radio galaxies matched in radio luminosity and redshift. Third, we note that the combination of a receding-torus-like scheme and the assumption that the observed submillimetre emission is dominated by quasar-heated dust yields a scaling relation L_{850} \propto L_{Bol}^{0.5} which is within the errors of that determined here for radio-selected quasars, and consistent with that inferred for radio-quiet quasars by Willott, Rawlings & Grimes (2003).