Indeed. When I saw a novel proposal for embedding QM inside GR using closed time-like curves (rather than the usual vice-versa), the main problem seemed to be that a GR-computer is potentially _much_ more powerful than a QM-computer, so you have to explain where the extra power has gone.