Time-delay strong lensing provides a unique way to directly measure the Hubble constant (H-0). The precision of the H-0 measurement depends on the uncertainties in the time-delay measurements, the mass distribution of the main deflector(s), and the mass distribution along the line of sight. Tie & Kochanek have proposed a new microlensing effect on time delays based on differential magnification of the coherent accretion disc variability of the lensed quasar. If real, this effect could significantly broaden the uncertainty on the time-delay measurements by up to 30 per cent for lens systems such as PG 1115+080, which have relatively short time delays and monitoring over several different epochs. In this paper we develop a new technique that uses the cosmological time-delay ratios and simulated microlensing maps within a Bayesian framework in order to limit the allowed combinations of microlensing delays and thus to lessen the uncertainties due to the proposed effect. We show that, under the assumption of Tie & Kochanek, the uncertainty on the time-delay distance (D-Delta t, which is proportional to 1/H-0) of the short time-delay (similar to 18 d) lens, PG 1115+080, increases from similar to 7 per cent to similar to 10 per cent by simultaneously fitting the three time-delay measurements from the three different data sets across 20 yr, while in the case of the long time-delay (similar to 90 d) lens, the microlensing effect on time delays is negligible as the uncertainty on D-Delta t of RXJ 1131-1231 only increases from similar to 2.5 per cent to similar to 2.6 per cent.