If I'm correct in my simplifying then all you need to do to reconstruct the past is to get a good model and a reconstruction of the past history of solar forcing and explosive volcanism (and some other things like land use). I may be taking your view to its extreme. I just cannot accept that any model will ever be as good as reality,
You mention ERA-40. This maybe a good reconstruction of weather over 40 years, but it is not a good reconstruction for climate purposes (see the attachment - this is an ERA-40 Report, a paper in JGR on the subject has been accepted). ERA-40 is better than NCEP is one conclusion from the work. Neither are good compared to the observations over the full 1958-2001 period. Most people who refer to Reanalyses use the word 'observations' and I always remind them they are not just simple observations.
ERA-40 is clearly very good for the period from 1979, but for the 1958-78 period there are clear problems because there are not enough observations going in to correct the model bias. It is even worse, ERA-40 rejected almost all Antarctic surface data before the late 1970s because it was too far away from the first guess field. This is partly because of weights ECMWF give to different sorts of data. From the late 1970s the satellite data are much more voluminous and the model finally begins to accept the surface data.The report might be of interest to a number on this email list. I think it proves that models still have a long way to go before they can be considered as alternatives to real observed data.
These Data Sets provide access to most of the data from the ERA-40
atmospheric model archived at ECMWF.