Predictive modelling of subsurface CO2 flow is used for optimising the design of geological CO2 storage projects e.g. with respect to mass stored and long term security. However several field scale projects have reported plume dynamics that do not match numerical predictions. Previous work has indicated that upscaling capillary heterogeneity results in different plume dynamics in synthetic 2D cross-sections and at early times (<0.2PV injected). Here we extend the workflow to 3D and analyse the impacts of longer term flow behaviour in an industrial scale geological carbon storage site with multi-point CO2 injection, structural relief and realistic flow rates. The models reveal that capillary heterogeneity in horizontally layered lithologies enhance channel formation, increases reservoir sweep, and reduces vertical migration speed near the injectors, all of which improve storage efficiency. At distances far from the injectors, the plume travels faster as a result of the heterogeneity. The results also show that simulations in 3D have qualitatively distinct results from those in 2D due to the increase in flow pathways.