Why?
Because its a very poor theory, with very poor backing, that relies on several large leaps of faith that ALL MUST be true, and even at that point doesnt really make sense.
First and foremost, Jim from chipworks who provided the die would have to be an incompetant buffoon, directly proven wrong by a random poster on gaf.
Than we would have to assume a random internet poster identified these parts because they have a similar (but not the same) geometric shape, without any knowledge of what alu's are inside. AND none of the over 50% of unidentified die space logic is rearranged registers unidentifiable because a computer customized the layout to maximize use of die space.
Then we have to assume his theory about a different fab process with renseas because they were the only supplier who could for some reason provide such undense edram of a different process size from the rest of the gpu (Which while technically possible... Has never been done that ive seen, anywhere)to make it so large (Nintendos switch to tsmc puts a serious wrench in this theory)
Rather than the common occurence of increasing density, and footprint to gain a desired amount, which is VERY common.