You can listen to me here (QuickTime Movie).
The problem with that test is partly that every measurement software you use is using IP packets. Depending on what efficiency of the software you can get the full link speed or not, for example depending on whether you use TCP or UDP. If you use TCP, then the window size, back off algorithms etc ensure that you are not using the link to 100%.
The test they have used uses not only TCP, but a plug in to browsers, so it is http that is used as transport.
Anyway, all of this ensure that you will never with that kind of software get the full speed of the link. So comparing is just not possible. What one can do is though a number of other things:
I think it is correct by the providers to say what throughput one can get with IP packets, but it should really be what I can get, with the most optimized software and operating system I have. I am not interested in theoretical throughput over the radio in some lab somewhere. IP itself “eats” some bandwidth. So the move the providers do that is described in the article is good.
What I can envision in the future is some normal usage scenarios with various applications that is to be used for tests. Similar to the specifications on gas usage for cars. But such definitions must be developed in cooperation with the ISPs.