But cant a chip like this one, be made into an ASIC version, yet keep doing what it does the way it does it? Its not based on the von Neumann bottleneck, so it should get to the solution differently (faster?)
Well all chips are "ASICs" however a general purpose pattern recognition chip isn't going to out perform a hyper optimized single purpose chip. Also there is no practical VN bottleneck in SHA-256d mining.