I don't really know enough about this to answer. In principle, though, if the sequences are short enough they might match those of other taxa just by chance. Also, I'm guessing here, because the collagen fufills a simple mechanical role, and therefore doesn't have to "co-evolve" in complex ways with other proteins, there would be more freedom for (noncrucial portions of) the protein sequence to evolve randomly in this direction and that, increasing a bit the likelihood that sequences from different taxa -- from different times -- might match by chance.
Both of the above are especially true of collagen isn't a particularly complex protein, which I suspect from it's structural nature, that it probably isn't.
That makes sense. But further reduces the value of what was discussed in the article. If the identity of the proteins was determined by chance, then there is just as much likelihood the seven protein fragments evaluated could have matched dolphins as chickens.