Folks...so far what I am seeing in this thread is bad science. (For the most part.)
You cannot take random tanks from different stations, different routes, approximate gas mileage and make accurate conclusions!
If you want to do this right... Drive 3,4, or even 5 tanks with one gas company driving the exact same routes (No special trips, massive changes in highway versus city miles) and average your MPG (Using M_Compact's method...he has it right) and then do the exact same thing with the other gas companies. Then come back and report the data and the conclusions... Otherwise, this is just opinion and conjecture.... and frankly, meaningless.