Mantle plumes are thought to recycle material from the Earth’s deep interior. One constraint on the nature and quantity of this recycled material comes from the observation of seismic discontinuities. The detection of the X-discontinuity beneath Hawaii, interpreted as the coesite-stishovite transition, requires the presence of at least 40% basalt. However, previous geodynamic models have predicted that the percentage of high-density basaltic material that mantle plumes can carry to the surface is no higher than 15–20%. We propose this contradiction can be resolved by taking into account the length scale of chemical heterogeneities. While previous modeling studies assumed mechanical mixing on length scales smaller than the model resolution, we here model basaltic heterogeneities with length scales of 30–40~km, allowing for their segregation relative to the pyrolitic background plume material. Our models show that larger basalt fractions than previously thought possible—exceeding 40%—can accumulate within plumes at the depth of the X-discontinuity. Two key mechanisms facilitate this process: (1) The random distribution of basaltic heterogeneities induces large temporal variations in the basalt fraction with cyclical highs and lows. (2) The high density contrast between basalt and pyrolite below the coesite-stishovite transition causes ponding and accumulation of basalt at that depth, an effect that only occurs for intermediate viscosities of pyrolite. These results further constrain the chemical composition of the Hawaiian plume. Beyond that, they provide a geodynamic mechanism that explains the seismologic detection of the X-discontinuity and highlights how recycled material is carried towards the surface.