I noticed a great deal of discussion of the random reconstruction here when I logged on wanting to let you know that a formal note has been published here:

Stockwell D.R.B., 2006. "Reconstruction of past climate using series with red noise", Australian Institute of Geoscientists Quarterly Newsletter, 83, pp14.

To be accurate this is not peer reviewed. Summary and links to the pdf can be found at my blog page http://landshape.org/enm/?p=30. I am sorry I have not followed the discussion here. From skimming it, it seems that I should highlight the interesting result is that 20% of the FARIMA series correlate with CRU temps, and 40% if you allow negative correlations as well (as MBH98 PCA methods do).

So the frequent correlation has to raise concerns with 'cherry picking' (which is largely due to LTP and short length of the CRU temperature series relative to variation). Steve McIntyre has continued to find high levels of within site variation in response temperature and 'missing' series.

Now the proportion of trees that correlate with temperature in a random sample at a site may be quite important I think, at identifying if the correlation of trees with temperature is entirely coincidental. But also the model of error is important as if you assume i.i.d. errors then virtually 0% would correlate with measured temperatues by chance, but in some other stochastic models, perhaps even more than 20%. So your assumptions are quite important as well.

But the bottom line is that the study does not claim to falsify the temperature reconstructions, just to point out that to use them as support for unprecedented 20th century increase and estimates of the MWP, is a circular argument. It is kind of trivial, its a trivial error, but what isn't trivial is the proportion of spurious correlations that should be expected when doing this kind of study.