When BMW dynos a car, they do it under controlled conditions in a temperature controlled environment, with no other motives than quality control.
When a random 3rd party dynos a car, they do it under questionable conditions, and questionable techniques.
As several people have already mentioned, it's very easy to add or remove horsepower from a car on any given dyno runs. I can take a stock E92 M3, and make more than one run on the same dyno on the same day, and get fluctuations high enough to sell just about any part I like.
Anyone who has substantial experience on a dyno can tell you that they can add 10hp to a car just by running the car differently, or setting up the correction factors differently.
I don't need to install an exhaust to produce a gain of 10hp or more. I can run our M3 bone stock once, and then run it again with slightly different setup and environment, and get the 10hp gain you'd like to see. Then I can post the two graphs of that stock M3 online, and say that our exhaust made those gains. Anyone can do that. Your system could have a net loss, and if you dyno it "right", you could show it to have a net gain.
Again, Eisenmann Germany's stance on this issue- as well as ours- reflects this fact.