http://falklumo.blogspot.com/2011/02/lumolabs-pentax-k-5-low-light-focus.html
Full test report / technical paper here:-
http://www.falklumo.com/lumolabs/articles/k5focus/index.html (in html format / online reading)
http://www.falklumo.com/lumolabs/articles/k5focus/K5Focus.pdf (in pdf format / offline reading)
I made a comment at Falk's blog previously, see:-
http://falklumo.blogspot.com/2011/02/lumolabs-pentax-k-5-low-light-focus.html?showComment=1298563006076#c5284422552664200059
And his responses:-
http://falklumo.blogspot.com/2011/02/lumolabs-pentax-k-5-low-light-focus.html?showComment=1298568951525#c6672564042501017507
http://falklumo.blogspot.com/2011/02/lumolabs-pentax-k-5-low-light-focus.html?showComment=1298569542930#c5929203098959810618
Whilst it is concluded that the AF reliability and accuracy of the K-5 / SAFOX IX+ could drop drastically for low light EVs, at or below EV8 (i.e., 1/60s at f/2.8 at ISO200, say for example), I found that the major limitation of his test is that the results are actually affected much by the optical defects of a lens. That says the test would yield more accurate and ideal results for a perfectly aligned lens (i.e., no de-centering of any) with perfect optical performance/design and deficiency (no CA and with very high central resolution, etc.).
With the de-centering defect that is commonly found in the Pentax lenses produced nowadays, similar FF or BF effects and thus deviation values could be derived as a result. See the following illustrations in Falk's technical paper:-
[Above: Back (Left) and Front Focused (Right) Images]
..which can be compared to the de-centered shot crops of the very similar chessboard chart taken by the Photozone earlier (which had all been deleted by Klaus anyway).
What I want to say is that Falk's method not only measure the AF errors, but also more so on the optical defects of a lens, which could be more noticeable and take more effects when the lens is wide opened.
Also, I think that mixing the results of various different lenses with which test shots were made under various conditions and present them in a single chart may not be a good enough idea. Yes, I think it is an "overall picture" that has certain reference value. But then presenting the test result lens by lens and one by one will surely reveal even more things and has even better reference value, I believe. Furthermore, to do a truly scientific and systematic test, the variable factor should be even limited to one at a time. I knew this would be extremely time-consuming and huge efforts are required, nonetheless.
Related:-
Magnified AF Errors of the SAFOX IX than the SAFOX VIII !?
SAFOX IX Tungsten/Yellow Light Front Focus Issue Fully Investigated