Re: However, the deniers will soon be along to point out ...
All the things I listed do falsify AGW. The tropospheric hotspot is mentioned as a predicted effect of AGW in the IPCC reports but it has not appeared at all. In fact, the upper troposphre where the hotspot is supposed to be has barely changed its temperature. If a theory makes a prediction and the opposite occurs, the theory is wrong.
The fact that CO2 and temperature don't correlate over the long term is in itself a huge falsification of AGW as it demonstrates there's no direct relationship between CO2 and temperature - if there's no relationship, there is no effect and no problem.
It only takes one of these things to demonstrate that AGW is wrong. We have two right there. There are plenty of others out there.
And on the subject of ensemble means: Weather forecasts average several runs of a single model, not several runs of several models. Given that a single model will tend to produce similar results with each run it is more likely to be "right" when you average it, but only for a given value of "right". After more than 3 days they are not very right at all and all the averaging in the world won't change that. You can get you position on a map "right" if you make a bunch of random dots and take the average to be where you are, but it's only "right" within a huge margin of error and if you start moving (changing over time as temperature does) the margin of error becomes so wide that it's functionally useless.
The ensemble mean of GCMs takes the averaged outcomes of several different models and averages them again. Given that each of these models does indeed miss out one or more major warming or cooling events in the 20th century, taking average of the averages of each models runs is also going to have such a wide margin of error as to be functionally useless. Each model gets it "right" in a very wide margin of error. The average get it "right" in an even wider margin of error. given we're talking about 10ths of a degree changes in temperature, and given these models diverge more than that from each other, taking the average seems to be a rather pointless exercise.