The hashing chips have yield problems (depending on the optimization of the density even 50% of chips can be bad). If You assemble a board with bad chips the whole board is bad (unless the board is designed for redundancy). On the other hand testing each chip separately is very expensive and time consuming (without special equipment that allows on wafer testing).
The only proof for the batch1 machines to be functional is a 50% increase of the hashing power.
I wrote many times already that SHA-2 is essentially self-testing and one of the cheapest possible to test. Anyone claiming otherwise is incompetent or deceiving.
I need to quote the signature too because it looks like the author is going to solicit money himself. It may be worth to keep eye on his business.
PicoStocks.com, High-tech startup fundraising, evaluation and profit sharing