It doesn't make sense to include `fast_run` if `fast_compile` mode is being used. Some rewrites such as the FusionOptimizer are not compatible with `fast_compile` mode which prevents the creation of C thunks. The FusionOptimizer has no way of knowing this is the case, and assumes it is safe to return Composites with more than 32 operands, even though that's not the case with the Python perform method.