Hi, statistician here.

For a single model to predict temperatures higher than what actually happens and for the result to be within the error bars is unremarkable. When *all 73* models predict results higher than what happens that indicates some serious systematic bias in the modelling (assuming there wasn't selection bias on the other side, of course--if the heartland folk intentionally ignored models that predicted less warming than actually happened then that's a huge problem). If the models are unbiased ("right on average") I would probably expect about half of them to predict above and half of them to predict below the actual results. (It wouldn't necessarily be exactly half and half, because the error distribution is not necessarily symmetric, but it should probably be somewhere around there.)

I believe that climate change is happening, but I think we're probably generally overestimating both the size of it and our precision.. There are well-recognized biases in various steps of the academic/scientific system--obtaining funding, getting published, making a name for yourself--that encourage this kind of exaggeration of results, in terms of both size and precision.

This is my judgement as a statistician--a kind of meta-scientist, if you will. I have no expertise in climate so I can't speak to the soundness of the mdoels being used, but the statistical behavior of them does raise some flags.