First off, by adding rapid shots you are simply subtracting from the numerator of a fraction (heat loss/time) thus making your ratio of those fractions (vanguard/merc) absurdly high. It simply doesn't help your cause when you inject math distortions like that and then trumpet the result, "Vanguards use 66.6% more resources than Mercs" which fails everyone's basic understanding of the matter. It's like claiming that wine has 10x the alcohol content of beer - if you first extract a fixed amount of alcohol from each fluid.
Secondly random lag doesn't affect both characters equally. It affects the mean (average) performance of each character equally. That is a statistics distinction which you may not be aware of. But your testing methodology is specifically vulnerable to random white noise, for reasons which I will try to explain. The basic segment of your test (FlameBurst + Rapid Shots) has almost NO measurable heat usage (1 unit). Over those two GCDs, your timing loss can easily be 0.2 seconds, which causes heat dissipation of - yes, 1 unit! Thus a common level of variance can completely disguise the very metric you are attempting to measure. That is why a test consisting of basic segments with a higher measurable heat usage, such as I used, are inherently more accurate.
Yes, you could get more accuracy by running the tests repeatedly. But using your methodology you might literally need to do it hundreds of times because the variance is so high relative to the mean of the metric you are measuring. Good luck with that.
Sometimes, no matter how hard both people try, it is not possible for the person taking the other side of an issue from you to prove that you are correct.
I did a manual run-through of the situation to eliminate any white noise. I put it in spoiler tags to save space (there may be errors). The assumptions I made:
0-39 high heat regeneration, 40 - 79 medium regeneration, 80 - 100 low regeneration.
12 - 8 high regeneration, 8 - 3 medium regeneration, 3 - 0 low regeneration.
As you can see, the Trooper gets 19 rotations. The Bounty Hunter gets 33.
33 > 19.
I'll assume the difference in values (from the real-world tests) is due to the white noise. There is still a substantial difference though. Feel free to point out anything that is inconsistent.