In 1998, there appears Climate Change Record in Subsurface Temperatures: A Global Perspective (Science 9 October 1998: 279-281) (subs req: sorry; abstract probably free) by Henry N. Pollack, Shaopeng Huang, Po-Yu Shen. The take-home message from that paper is pretty much the graph from http://www.ncdc.noaa.gov/paleo/borehole/core.html, which I’ll inline here (and its the same as used in the TAR). And that is: a 500y record, showing a temperature rise in line with the instrumental record where they overlap (and somewhat colder than most other reconstructions around 1500; but thats not the point here; if you’re interested try this by Mann at al. which reconciles them). And Beltrami 2002 gets a similar-ish result. So that appers to be the current situation with borehole stuff: it goes back 500 years.
But, the point is that the same people (HP&S, but in a different order) a year earlier published in GRL (SR, of course :-() a reconstruction going back 20,000 years. There are features of that that appear distinctly odd to me – they only get about 1.5 oC cooling at the last glacial maximum; and to get more it looks like they would have to increase temperatures in the holocene too high. And the LGM was… at least 5 oC colder. Also the timings are wrong: the “node” between the “LIA” dip and “MWP” peak is at 500 years before present, which is when the paper a year later ends with maximum cooling.
And its this earlier paper that McK picks up to argue for a MWP.
So why has the earlier paper disappeared? Of course the true explanation must be nefarious interference from the UN/IPCC to suppress the truth; but what is the excuse? I’m not really sure, and would be grateful for anyone who does know. The 1997 GRL paper appears to use heat flow and about 6000 sites; the 1998 Science uses temperatures and 358 sites. So maybe it turned out there was something badly wrong with the earlier method? It certainly contradicts the later one; and produces implausible values for the LGM temperatures.
The paper “only” gets referenced 17 times, according to WoS, and always in a “various people have done boreholes, including Huang (1997)” sort of way. I can’t see anyone who has used or commented on their particular profiles (apart from the recent uptake by the skeptics, of course).
[Update: Eli has already noticed this oddity (and there is more) but he doesn't explain it either. He does discover Pollack and Huang in 2000 referencing both studies but apparently failing to notice their incompatibility -W]
[Another update: I've had some email exchanges with a very friendly Henry Pollack. He supplies almost all the answers: part of it is HPS 97 paper had data to depths of two kilometers, whereas the 500 year database comprises boreholes mostly in the 300-400 meter depth range. The much larger number of boreholes in the HPS 97 dataset did enable the greater depths to be sufficiently populated to have a reasonable estimates of the mean heat flux over a depth interval. OK, so this is why HPS '97 goes to 20 kyr and all the subsequent ones only to 500 y. But then this leaves the central problem: why is a larger database of deeper boreholes no longer used? The answer to this seems to be Quality. The datasets used in the 500y studies are better controlled. The HPS '97 dataset is of heatflow, derived from the International Heat Flow Commission database. But though the heatflow measurements there were derived from temperature measurements, the original Ts are gone and only the heatflow is in the database.
An interesting additional point is that since HPS didn't use the top 100m of borehole, the reconstruction there contain virtually no information about the 20th century... the 'present' (the zero on the time axis) really represents something like the end of the 19th century, rather than the end of the 20th century... the present-day is indeed warmer than the 'goldilocks' curve b (as well as curve a) throughout the Holocene, and at least as warm as the Medieval Warm Period of curve c (quotes from Pollack).
Which is brings us back to the original point: no-one now uses HPS '97. The shorter record is preferred, as (presumably) being more reliable -W
ps: datestamp adjusted to push to the top]