The superfluid drag-coefficient of a weakly interacting three-component Bose-Einstein condensate is computed on a square optical lattice deep in the superfluid phase, starting from a Bose-Hubbard model with component-conserving, on-site interactions and nearest-neighbor hopping. At the meanfield level, Rayleigh-Schrödinger perturbation theory is employed to provide an analytic expression for the drag density. In addition, the Hamiltonian is diagonalized numerically to compute the drag within mean-field theory at both zero and finite temperatures to all orders in inter-component interactions. Moreover, path integral Monte Carlo simulations, providing results beyond mean-field theory, have been performed to support the mean-field results. In the two-component case the drag increases monotonically with the magnitude of the inter-component interaction γAB between the two components A and B. The increase is independent of the sign of the inter-component interaction. This no longer holds when an additional third component C is included. Instead of increasing monotonically, the drag can either be strengthened or weakened depending on the details of the interaction strengths, for weak and moderately strong interactions. The general picture is that the drag-coefficient between component A and B is a non-monotonic function of the intercomponent interaction strength γAC between A and a third component C. For weak γAC compared to the direct interaction γAB between A and B, the drag-coefficient between A and B can decrease, contrary to what one naively would expect. When γAC is strong compared to γAB, the drag between A and B increases with increasing γAC , as one would naively expect. We attribute the subtle reduction of ρ d,AB with increasing γAC , which has no counterpart in the two-component case, to a renormalization of the inter-component scattering vertex γAB via intermediate excited states of the third condensate C. We briefly comment on how this generalizes to systems with more than three components.