Mutualisms involve cooperation between species and underpin several ecosystem functions. However, there is also conflict between mutualists, because their interests are not perfectly aligned. In addition, most mutualisms are exploited by parasites. Here, we study the interplay between cooperation, conflict and parasitism in the mutualism between fig trees and their pollinator wasps. Conflict occurs because each fig ovary can nurture either one seed or one pollinator offspring and, while fig trees benefit directly from seeds and pollinator offspring (pollen vectors), pollinators only benefit directly from pollinator offspring. The mechanism(s) of conflict resolution is debated, but must explain the widespread observation that pollinators develop in inner, and seeds in outer, layers of fig flowers. We recently suggested a role for non-pollinating figs wasps (NPFWs) that are natural enemies or competitors of the pollinators and lay their eggs through the fig wall. Most NPFW offspring develop in outer and middle layer flowers, suggesting that inner flowers provide enemy-free space for pollinator offspring. Here, we test the hypothesis that NPFWs cannot reach inner flowers, by measuring wasp and fig morphology at the species-specific times of NPFW attack in the field. We found that three species of Sycoscapter and Philotrypesis wasps that parasitise pollinators could reach 34–73%, 75–92% and 82–97% of fig ovaries, respectively. Meanwhile, Eukobelea and Pseudidarnes gall-formers, despite having shorter ovipositors, can access almost all fig flowers (93–99% and 100%), because they attack smaller (younger) fig fruits. Our mechanistic results from ovipositing wasps support spatial patterns of wasp offspring segregation within figs to suggest that inner ovules provide enemy-free-space for pollinators. This may contribute to mutualism stability by helping select for pollinators to avoid laying eggs where they are likely to be parasitised. These outer flowers then remain free to develop as seeds, promoting mutualism persistence.