Predators regulate communities through top-down control in many ecosystems. Because most studies of top-down control last less than a year and focus on only a subset of the community, they may miss predator effects that manifest at longer timescales or across whole food webs. In southeastern US salt marshes, short-term and small-scale experiments indicate that nektonic predators (e.g., blue crab, fish, terrapins) facilitate the foundational grass, Spartina alterniflora, by consuming herbivorous snails and crabs. To test both how nekton affect marsh processes when the entire animal community is present, and how prior results scale over time, we conducted a 3-year nekton exclusion experiment in a Georgia salt marsh using replicated 19.6 m2 plots. Our nekton exclusions increased densities of plant-grazing snails and juvenile deposit-feeding fiddler crab and, in Year 2, reduced predation on tethered juvenile snails, indicating that nektonic predators control these key macroinvertebrates. However, in Year 3, densities of mesopredatory benthic mud crabs increased threefold in nekton exclusions, erasing the tethered snails' predation refuge. Nekton exclusion had no effect on Spartina biomass, likely because the observed mesopredator release suppressed grazing snail densities and elevated densities of fiddler crabs, whose burrowing alleviates soil stresses. Structural equation modeling supported the hypotheses that nektonic predators and mesopredators control invertebrate communities, with nektonic predators having stronger total effects on Spartina than mud crabs by controlling densities of species that both suppress (grazers) and facilitate (fiddler crabs) plant growth. These findings highlight that salt marshes can be resilient to multiyear reductions in nektonic predators if mesopredators are present and that multiple pathways of trophic control manifest in different ways over time to mediate community dynamics. These results highlight that larger scale and longer-term experiments can illuminate community dynamics not previously understood, even in well-studied ecosystems such as salt marshes.