I had assumed that it must be Turing complete because otherwise its utility as an intermediate language seems dramatically reduced (despite its power, no one is compiling to Coq!); but, now that you point it out, I agree that there is no such claim. I originally thought that general recursion followed from folds and unfolds (http://lambda-the-ultimate.org/node/4290), but Ross Angle points out at http://lambda-the-ultimate.org/node/5045#comment-82340 that that's only when they are mixed, which is apparently not the case here.