ABSTRACT A recent measurement of the Lyman-limit mean free path at z = 6 suggests it may have been very short, motivating a better understanding of the role that ionizing photon sinks played in reionization. Accurately modelling the sinks in reionization simulations is challenging because of the large dynamic range required if ∼104−108M⊙ gas structures contributed significant opacity. Thus, there is no consensus on how important the sinks were in shaping reionization’s morphology. We address this question with a recently developed radiative transfer code that includes a dynamical sub-grid model for the sinks based on radiative hydrodynamics simulations. Compared to assuming a fully pressure-smoothed intergalactic medium, our dynamical treatment reduces ionized bubble sizes by $10-20~{{\ \rm per\ cent}}$ under typical assumptions about reionization’s sources. Near reionization’s midpoint, the 21 cm power at k ∼ 0.1 hMpc−1 is similarly reduced. These effects are more modest than the $30-60~{{\ \rm per\ cent}}$ suppression resulting from the higher recombination rate if pressure smoothing is neglected entirely. Whether the sinks played a significant role in reionization’s morphology depends on the nature of its sources. For example, if reionization was driven by bright (MUV < −17) galaxies, the sinks reduce the large-scale 21 cm power by at most 20 per cent, even if pressure smoothing is neglected. Conveniently, when bright sources contribute significantly, the morphology in our dynamical treatment can be reproduced accurately with a uniform sub-grid clumping factor that yields the same ionizing photon budget. By contrast, if MUV ∼ −13 galaxies drove reionization, the uniform clumping model can err by up to 40 per cent.