In the first article Stuart pointed out that the major areas of consumer spending seem not to have changed much in the past 20 years, and that explanations for improvement in "barrels per dollar" of GDP come in two varieties: a shift in goods and services to things that don't depend on oil, or more efficient use of oil. Two other options were raised later and in comments: the measurement of "constant dollars" is off, somehow, and more significantly, the measurement of oil use is off, caused by globalization of energy-intense industry to other countries.
The globalization problem definitely complicates calculation of energy intensity averages, as we've discussed here before. Simple scenarios can have energy intensity falling within all the individual nations, and yet rising worldwide if the mix of economic activity shifts from more efficient to less efficient producers.
In Staniford's latest, there's also a possible problem with averaging, as I pointed out there: Even if the weighting is by highway miles driven, it still might not give you the right improvement ratio.
For an example, let's say cars get 20 mpg, light trucks 15 mpg, and other trucks 5 mpg. Let's say of every highway mile driven, 40% are cars, 40% light trucks, and 20% heavy trucks. Then, with those percentages you get a "weighted average" mpg of 15 (0.4*20 + 0.4*15 + 0.2*5). However, the actual number of gallons of oil used for one "average" mile is 0.08666 (0.2/5 + 0.4/15 + 0.4/20) giving a true average mpg of 11.5, not 15.
I.e. the real improvement may be much less than you would get from a simple weighted average of sector mpg's.