I dunno, spend a couple of weeks out of the country and a proper debate strikes up for a change. Marotzke and Forster published this paper basically arguing that the AR5 models don't generally either over- or underestimate temperature changes, based on various trends within the last century or so (up to now). My first thought on a superficial glance at the paper was that it wasn't really that useful an analysis, as we already know that the models provide a decent hindcast of 20th century temps, so it's hardly surprising that looking at shorter trends will show the models agreeing on average over shorter trends too (since the full time series is merely the sum of shorter pieces). That leaves unasked the important question of how much the models have been tuned to reproduce the 20th century trend, and whether the recent divergence is the early signs of a problem or not. (Note that on the question of tuning, this is not even something that all modellers would have to be aware of, so honestly saying "we didn't do that" does not answer the question. But I digress.)
One limitation of the MF study was that they do not know the forcing for each model, or the α and κ parameters, so had to estimate them by linear regression based on (among other things) their temperature time series. Along comes Nic Lewis, and says "aha - this is circular". Now I haven't had time to look into it in detail, but this argument clearly has some validity in principle. MF replied here, but I'm not really impressed by what they said. There is a certain amount of talking past each other - MF are saying that their method is physically reasonable (which it is) and, in previous work, gives good results, however they don't really address the main criticism. Lewis says their method is simply invalid, because the assumptions underlying the statistical theory are violated. In that respect, it is worth pointing out that probably just about all statistical analysis is simply invalid at some level, since the assumptions are rarely precisely correct. Exact linearity, independent errors, gaussian statistics? You've got to be joking. These are never more than approximations to the truth, but hopefully the approximations are good enough that the end result is useful.
Some of the commenters on the climate lab book thread seem to have got a good handle on it. Just to expand on it a bit, if MF got the "correct" values for forcing, α and κ then it wouldn't matter where these numbers came from. However, there will be some uncertainty/inaccuracy in the estimates they have derived, and these did come from the temperature time series, and will lead to some circularity. So the question is really whether these inacuracies are big enough to matter. I have an idea for investigating the magnitude of the problem (which a priori could either be negligible or could indeed invalidate the work). I'm not sure it's really worth the effort though.