If the core speed were ever increased, it would be as a side-effect of using a much smaller process node. Maintaining the original clock speed gets tricky as you continue reducing the process node for a fixed design due to various transistor parameters, especially if you want to also reduce power consumption. There's a balance to be made between clock cycle, operating voltage, and threshold voltage ultimately figuring into static and dynamic power.
The reduction in operating voltage and thus active power consumption is a necessity born out of how transistors are made and function; Vthreshold is the culprit. It would not be good if a particular circuit of transistors were always turned on or off because the threshold was always exceeded. That's not to say you couldn't intentionally design for low power, but the threshold voltage is usually reduced with a smaller transistor, leading to an exponential rise in static/idle power consumption; this is rather counterproductive for thermal density. Some solutions to the Vth issue may be different gate materials or different dopant concentrations or even different transistor design altogether, all of which open even more cans of worms.
If the situation with leakage is dire (i.e. can't do anything about it), 1) don't bother or 2) increase clock frequency because the leakage current and parasitic capacitances cause an increased discharge rate at the output of the transistor, resulting in errors; as the voltage supply is decreased, the system will be even more sensitive to fluctuations. The higher frequency will help keep the output at what it should be.
Temps is also not really the indicator of why it may have been changed, it most likely was yields and power consumption as well as not being able to provide a higher stepping/core revision probably due to not being able to simply throw out undesirable GPU wafers.
Temperature/heat dissipation is related to yields. Recall, not all chips are equal hence the binning often seen in the PC side. But yes, considering the launch of G71in March 2006, it's not too surprising; the highest grade @ stock 650MHz required a hell of a heat heatsink and fan combination in a much more open chassis . Even the GT/GS @ 450 had rather hot single-slot solutions.
