[1] Fundamentally contradictory interpretations of the isotopic compositions of Hawaiian basalts persist, even among authors who agree that the Hawaiian hotspot is caused by a deep-mantle plume. One view holds that the regional isotopic pattern of the volcanoes reflects large-scale heterogeneities in the basal thermal boundary layer of the mantle. These are drawn into the rising plume conduit, where they are vertically stretched and ultimately sampled by volcanoes. The alternative view is that the plume resembles a ''uniformly heterogeneous plum pudding,'' with fertile plums of pyroxenite and/or enriched peridotite scattered in a matrix of more refractory peridotite. In a rising plume, the plums melt before the matrix, and the final melt composition is controlled significantly by the bulk melt fraction. Here we show that the uniformly heterogeneous plum pudding model is inconsistent with several geochemical observations: (1) the relative melt fractions inferred from La/Yb ratios in shield-stage basalts of the two parallel (Kea-and Loa-) volcanic chains, (2) the systematic Pb-isotopic differences between the chains, and the absence of such differences between shield and postshield phases, (3) the systematic shift to uniformly depleted Nd-isotopic compositions during rejuvenated volcanism. We extend our previous numerical simulation to the low melt production rates calculated far downstream (200-400 km) from shield volcanism. Part of these melts, feeding rejuvenated volcanism, are formed at pressures of $5 GPa in the previously unmelted underside of the plume, from material that originally constituted the uppermost part of the thermal boundary layer at the base of the mantle.