Mutualism can evolve when organisms make novel connections that happen to benefit all parties. When such connections involve a host that provides a public good for multiple visitor species, selection for reciprocal cooperation is only likely if the host has the power to preferentially increase the fitness of those visitors that provide a better quality service. In contrast, when interactions form between multiple visitors and a host that lacks the power to partition the public good, we predict that interactions are likely to persist only as by-product mutualisms, in which symbiosis benefits both host and visitor, but in which neither partner makes costly investments. Focusing on the symbiosis between ants and litter trapping epiphytes in the canopy of lowland dipterocarp rain forest in Borneo, we show that at least 71 ant species share the public good of housing within the root-mass of epiphytic bird's nest ferns (Asplenium spp.). Ferns supporting a higher biomass of ants experienced less herbivory, and experimental exclusions of ants from fronds confirmed that this is caused by the ants protecting the ferns from herbivores. These results establish that there are clear by-product benefits for both parties of housing for ants and protection for ferns. If these benefits were to drive selection for reciprocal cooperation, we would expect larger ferns to support ant colonies that were larger or colonies that produce more reproductive individuals. This was not the case. Larger ferns instead supported more ant colonies, indicating that the housing provided by the fern is a public good that is not restricted to better cooperating ant species. Mutualism via novel connections is thus unlikely to evolve between a host and multiple partners, even if there are clear by-product benefits to all parties, unless the host can direct benefits to cooperators.