This is about “Model projections of the North Atlantic thermohaline circulation for the 21st century assessed by observations” by A. Schmittner, M. Latif, and B. Schneider (here if you have a GRL subscription). Its interesting for two reasons: another data point on the “will the THC shutdown” (no); and because of the way it weights multiple AR4 models.

What they are interested in is the size and probability of a THC slowdown over the coming century. To do this they look at 9 models (some with multiple realisations) from the AR4 archive under SRES A1B. The end result is a weakening of the THC from about 16 to about 12 Sv. They don’t say what that does to the temperature, but I think it still leaves warming over Europe.

More interesting to me is the weighting thing: they compare various variables in the models to obs climatology, form a skill score, and then weight the models contribution to the final result by W=exp(-0.5S^2). Weights then vary from 0.04 (ie, almost thrown out) to 2.75 (used). This is a Good Thing to do, because some of the models are not really very good, and just averaging them unweighted fails to reflect this. OTOH, they do point out that the mean result is pretty similar, weighted or not, though the uncertainty range is less for the weighted models.