The author seems to be missing a bit of Linus' point. It seems that the point Linus is making is that a collision along with the underlying data being meaningful is orders of magnitude harder than the collission, or produces easily detectable abberations. I'm not a cryptographic expert either but the author seems to be thinking more in a world of passwords where the underlying data often remains unknown. I think Linus is being somewhat pragmatic too and basically saying that if you think this is a problem then the whole thing is a house of cards.