Ant-plants are important structural elements in many disturbed tropical ecosystems and the mutualism between plants and their defending ant symbionts is increasingly being used as a model to study general factors that stabilize a horizontally transmitted mutualisms. As these mutualisms must be established anew in every consecutive generation they are particularly prone to the destabilization by cheaters (former mutualists that ceased the service provisioning) or parasites (non-reciprocating exploiters with no evolutionary history as a mutualist). Theoretical models predicted high exploitation rates for high-reward mutualisms. We empirically tested this prediction and found that sympatric Mexican Acacia myrmecophytes differ at the species level in the amount of food rewards and nesting space that they provide to their defending Pseudomyrmex ant mutualists. Several biochemical specializations help to exclude non-adapted potential consumers from feeding on the host-plant derived food rewards: food bodies and extrafloral nectar are biochemically protected from nectar-infecting microorganisms, nectar robbers and herbivores. Interestingly, the same adaptations appear to make the symbiotic ants fully dependent on these rewards. At the phenotypic level, hosts that produced more extrafloral nectar were more aggressively defended by their inhabitants. Genetically fixed, species-specific plant traits combine with phenotypic plasticity to create 'host sanction' mechanisms, which bind fitness-relevant traits of both partners to each other. This assumption is confirmed by the observation that two high-reward producing among four investigated host species were less commonly exploited by non-defending ant species. As reward production can be costly, this allows the diversification into 'low cost-high risk' versus 'high cost-low risk' strategies, and both strategies are indeed realized by sympatric and congeneric host species.