Popular Mechanics does some good consistent testing on gas mileage improvement machines. I'd suggest checking it out.

I am familiar with Mike Allen's evaluations. He admits that they are not "scientific". About the closest thing to a "scientific" evaluation of vehicle fuel efficiency of heavy diesel vehicles (Class 3-8) is the SAE J1321 standard. One J1321 evaluation of an oxyhydrogen system is found here. This was done on very heavy trucks (over 100,000 lbs.) designed to perform wear tests on pavement. Performance of these systems varies widely with speed and load. On these tests, improvement was rather marginal. Better results were obtained with this evaluation done under simulated highway and city driving conditions with two cabs minus their trailers. Although it did not follow the J1321 protocol, I have the raw data and I found that improvement was statistically significant by a wide margin. Also, meteorological data was recorded and the pattern of the data taken over a period of about 1 month indicates that the difference is not likely to be weather related. The original version of J1321 left a lot to be desired. It actually was not very rigorous and made no recommendation regarding statistical analysis of data. It was not until Feb. 2012 that a revised version was released that corrected this shortcoming. As of this date, I do not know of any evaluations done following the revised J1321 standard. As I said, the technology needs to be better validated.