Bill.
Your graph is reminiscent of the ancient ""Radiate Calculated"CPU Load v "In Socket" DeltaTcpu - water " compilation.
However, in the light of the work by
Incoherent and
pHaestus , I tend to think that 100% Radiate(type) calculated Wattages maybe closer than 70%..
Not sure,but think your graph represents an "Insulated Die"(Ri=15 c/w) . Would suggest that the graph for a "Board-mounted Die"(Ri=1 ish?) would have a similar gradient to that for a CPU(using Radiate(type) Wattage).
An illustration of Kryotherm predictions: