Now, i presume what he has said is based on this data:
http://spreadsheets.google.com/pub?key=0AtqsF2x2K22ocHltMHZWd1ZIZFNVRUpJSXpQbks3ZUE&gid=1
However, i believe this data to be disputable, mainly because, although from what i have seen the test has been carried out as fairly as possible, there are still outliers in the data as calculated from the LB = Q1 - IQR and UB = Q3 + IQR method.
I have identified a few outliers to test this theory however i will calculate it for .683 (of which they claim is the proof of underboring being more accurate.
EDIT: Suprisingly, i found no outliers in .683 and i must submit to conclude that according to the data, you are correct, there is a slight improvement in accuracy with underboring.
However, i still stick to the science over the maths in that i think bore matching still makes a difference. If you plot the figures into a graph, you can see there is a noticeable increase in consistency towards the ball bore (.685) but with .683 taking precedence as the lowest.