It seems to me the essential features of the Penrose tiling are that (a) there are only a finite set of local patterns (b) the Penrose tiling can be got from a higher dimensional lattice by considering a 2D subspace of the higher space, and projecting stuff that is near to that subspace, orthogonally into the subspace. Due to (a) you can hope to devise a CA and write down only a finite number of rules to define it. Due to (b) there is some notion of "direction" and what physicists call "long-range order." (One way in which that is revealed is the Fourier transform of Penrose, which has delta functions like in diffraction from a crystal. Another is "Ammann bars.") Indeed one way Goucher could have worked (although maybe not the way he actually did it) might be to devise a CA on the higher-D lattice whose rules still worked restricted to the Penrose subset alone. If instead we were to use the Voronoi diagram of Poisson-random points in the plane, then I doubt Goucher (or anyone) could make a CA on THAT with gliders that keep going in some direction... due to properties (a) and (b) now failing...