That's been tested in court before, in Sega v. Accolade (http://en.wikipedia.org/wiki/Sega_v._Accolade). (You had to display a Sega logo before your game could run, so you had to have their logo in your code)
In short, it's fair use.
My understanding from the Wikipedia article you linked is that you just had to have the string SEGA in a specific spot on your ROM, and that the logo was in fact part of the Genesis III BIOS. A four-character string and an image are pretty different, copyright-wise.
You publish a benchmark that says how inefficient this makes MySQL, and then that "feature" goes away.
I was more concerned about the copyright repercussions on clean-room reverse engineering the protocol. Obviously the protocol includes copyrightable information, now.
To the best of my knowledge, the best answer is "nobody knows". There are some vaguely similar things in the precedents, especially when it comes to protocols that require you to use a particular patent to talk to a given device, and IIRC there is some verbiage in some places that distinguishes copying around for simple transmission reasons rather than delivery to an end user (though I can't imagine how to stretch that to cover this case), but in this particular case, nobody knows.
In fact if you spend much time thinking about this it isn't at all hard to come up with a lot of scenarios that law doesn't have an answer for, in theory or in practice.
What would happen, if the protocol would require sending a picture of a picture of Mickey Mouse with each request (or some copy-righted novel)?