IBM does provide some new G5 power data in a recent research article, titled Power Measurement on the Apple Power Mac G5.
Using a series of different benchmarks with a dual G5 970 2.0 Power Mac, the article provides various power usage ratings at different levels of CPU utilization. We see a maximum power dissipation of 79 Watts, with the MS_hot benchmark, which "executes exclusively out of the L1 cache, performing a tight loop of square root instructions".
As I mentioned in a previous article, the published typical power utilization of the G5 970 1.8 GHz (130 nm) is 51 Watts, and the maximum power has been estimated by some at 90+ Watts. (IBM doesn't usually publish maximum power ratings for the G5 for public consumption.) Given that the G5 970 2.0 GHz has a maximum power of 66 Watts, one might estimate the maximum power at well over 100 Watts. However, this IBM research article disputes those estimates, and measures the maximum power dissipation of the G5 970 2.0 at 79 Watts.
The truth may lie somewhere in between. The benchmark likely does not maximize power utilization completely, but it is possible the estimates of 100+ Watts for the G5 970 2.0 and 90+ Watts for the G5 970 1.8 may be too high. However, even if we were to assume the effective maximum power of the G5 970 1.8 is say 85 Watts (which is higher than the measured 79 Watts for the faster 2.0 GHz CPU), then we might guess that the effective maximum power of the G5 970FX 2.5 is also close to 85 Watts (since the typical power of the 970FX 2.5 is almost identical to the 970 1.8).
Extrapolating, one might guess that the G5 970FX 3.0 at the same voltage (as the 970FX 2.5) might come in at just over 100 Watts. And even if the voltage for the 3.0 needed to be boosted, it still could come in at the 110 Watt range, which although hot, is still quite acceptable for a desktop CPU.