Can you prove an upper bound on parallelization? In other words, can you prove that the advantage of x working units as x->oo goes to zero? Empirical evidence about how hard people have found parallelization of factoring to be is just hand-waving.
Once you prove that SHA-256 encryption can't be broken without using brute force. You trying to draw this hard line of definitive proof shows that you don't have the wisdom to understand how the world works.
You're shifting the goalposts. Your opening claim was that factorization does not benefit from GPUs (by which, I assume you mean parallelization). I nitpicked by pointing out that there's no reason to believe that factorization can't be sped up by parallelization. Claiming that factorization cannot benefit from parallelization is tantamount to claiming that there is no linear
speedup for factorization. If this were true, it could be proved. There is no other way to know that it is the case than to prove it.