As best I can tell it is impractical as an actual decision-making procedure for more complex cases, at least assuming well-formalized priors. As a limit to be asymptotically approached it seems sound, though -- and that's probably the best we can do on our hardware anyway.