Ok, I think this is officially a false alarm, based on more testing by Jim Titt. See the threads linked above.
Yes and no.
It is a false alarm in one sense: it has been clearly shown that with the Yosemite finish in it, the bowline is quite safe even if it has been subjected to this pulling on the "wrong" strand.
But it is not a false alarm in the sense that we have learned that there is a method, now documented, by which the Yosemite finish, when tied just so, and when the wrong strand is pulled, will modify the bowline, such that if the Yosemite finish is removed, the bowline no longer remains.
I do think the latter is worth noting, even if it's somewhat academic. Hard to know when it might come into play, but still well worth adding to the knowledge-base for future reference.