Well, the climate model simulations are not done to predict the exact position of features around the globe in the future. Rather, they are to try to distinguish what differences may occur in the relative frequency of extreme events and changes within the mean of many parameters (like area-averaged temperatures) by adding the radiative forcing of increasing greenhouse gas concentrations compared to model runs that do not have increasing greenhouse gas concentrations. The idea is to separate natural variability from the anthropogenic forcing.
The patterns that you are looking for in terms of everyday/weekly weather patterns are considered no more than noise due to natural variability in such an analysis. I believe the study you cited is looking at the frequency of occurrence of extremely hot summers, and so while not every summer in the 2080's in the model may be extremely hot, it is more likely compared to a year in earlier in the model run.
To be honest, I don't know what you mean by "past-cast mode". If you mean the pre-industrial control runs, it is hard to verify, since you would essentially be relying upon paleo-data. The models do show relatively stable control runs, with no significant climate drift, though. The recent IPCC AOGCM twentieth century experiments (which follow the anthropogenic emissions observed in the twentieth century and include aerosol and volcanic forcings) do capture the trend and variability of area-averaged temperature quite well compared to the observed record in the US, and globally. They do not replicate major climate events in the US at the same time as they were observed, but do have the capability of reproducing certain characterstics (such as the warmth of the 1930's, but not the magnitude of the drought of the 1930's) at any given time.
I agree with Derek; that while you may not be able to pinpoint the exact setup in the future, you can look at changes in the mean conditions or in the frequency of extreme conditions in a long-term sense.
