I've always run 91. My figuring is that the increased octane lets it run more timing across the board, given that it has an octane sensor and apparently makes significant use of it.
More timing is more efficient. Both 87 and 91 contain the same amount of energy (i.e., "BTUs"), but with 91 you can apply that energy better by starting the combustion of the fuel with the piston at a more efficient position. With more timing, you can get more rotational energy down the crankshaft for a given amount of fuel.
I could be wrong in this, the timing could be the same for 87/91 at non-WOT/closed loop. It could also be that the ideal timing is the same with 87 and 91 at the lower-load conditions when one is driving economically.
I thought I observed a slight increase in MPG with 91 as well, but that wasn't exactly scientific, and like every such non-controlled test, has a huge chance of being grossly inaccurate. It being a bit warmer or windier on one test versus the other would invalidate the results.
