Context <p>Marine reserve networks designed to enhance larval dispersal provide important biodiversity benefits. Designs are commonly based on time-averaged means of dispersal estimates. It is unclear whether they capture the connectivity portfolio effect, by which temporal variation in individual reserve performance is buffered by the entire network.</p> Objectives <p>To evaluate the implications of dispersal variability and derive general rules for network design.</p> Methods <p>We modelled larval dispersal of four commercially important reef species for 20&#xa0;years in Southeast Sulawesi, Indonesia. We built reserve network scenarios informed by temporal dispersal as either an interannual mean or dispersal for individual years. We then evaluated reserve network performance.</p> Results <p><i>Plectropomus leopardus</i> and <i>Siganus canaliculatus</i> experienced higher variability in interannual larval flow compared to <i>Lutjanus malabaricus</i> and <i>Octopus vulgaris</i>. We discover that using a single time-averaged mean dispersal estimate can create subpar reserve networks in highly variable systems. Mean dispersal was suitable for less variable species such as <i>O. vulgaris,</i> but not for more variable species where high larval flow was not protected over time. By explicitly identifying reefs contributing a high amount of larval flow in each of the 20&#xa0;years, we were able to improve network designs and provide a more consistent protection of larval supply over time.</p> Conclusions <p>We develop methods to improve network performance by protecting reefs with high larval flow contribution across years. The steps outlined here illustrate how information from multitemporal connectivity datasets can help inform a spatial prioritisation framework to accommodate larval dispersal volatility.</p>