There is a seriously interesting question here about the information content of the theorem and whether it is equivalent to the Shannon limit on compressing the

There is a seriously interesting question here about the information content of the theorem and whether it is equivalent to the Shannon limit on compressing the 'code' needed to run the computation. I bet someone published on this years ago but I haven't seen it phrased in quite the way you have it in point 2.