I must repeat one of my arguments.
PowerPC 970FX. 58 milion transistors. 10GFLOPS. 33.5 milions used for logic (not cache).
PowerPC 750VX 1.5Ghz. 10.5 milion transistor (is the IBM equivalent of the PowerPC 7410) without the L2 Cache. 9 GFLOPS.
Now take the Gekko (the original G3 is 6.35 milion transistors) and you will see that the logical part without L2 Cache is 9 milion transistors, if we add the VMX unit the thing must go to the 13 milion transistors and the number of FLOPS will go from the 9 GFLOPS to the 12GFLOPS, a 20% more floating point power than the solution of using a PowerPC 970FX and with the advantage of using less transistors.
The other part is the 1T-SRAM in the L2 Cache, normally the L2 cache is 6T-SRAM and this has an huge impact on the heat consumption of the processor, if we reduce the number of transistors needed for making a work we will have a more "fresh" CPU or we can going to high clock speeds.
My main idea dual Gekko+VMX with multithreading support, and a Dual Gekko+VMX+MT has in theory less transistors than a PowerPC 970FX in the logical part, after it you use 1T-SRAM and you obtain my theory of the Revolution CPU.