If it is the same thing I ran into (which I believe it is), then the routine that does it is one which supposedly works out whether certain memory spaces are mirrored (which is why you don't see any reads --- the reads are at the unmirrored position); IIRC correctly, the routine had always been there, but for BB2, H&P made it go just a bit further than before.
Now, the question of "why" is interesting. While I am of course not party to any decision making at H&P, I have a suspicion. Before releasing BB2 on the unsuspecting public, H&P had just screwed up the whole Amithlon thing in an impressive manner. They were aware that a successor product was in the pipeline, and that they would not be involved in it. In fact, as a matter of courtesy, I had made an early version of the successor product available to them, with the request to please check and tell me about any specific IP of theirs they considered to be present in it.
When BB2 came out, it broke what later was dubbed "Umilator". The reason being that Umilator, unlike Amithlon or any real hardware Amiga, used that space to map RAM into it. And it so happened that some of the updates got loaded into it, then got corrupted (because the silly routine did its tests destructively). Uh-oh!
I would prefer to think of this as an innocent mistake, rather than an (ultimately futile, of course) act of sabotage --- but in order to think so, it would be very helpful to find even a single piece of hardware that the changed test has any use on. So far, no dice.
