> Shouldn't this be a requirement for all patents?

It has to demonstrate utility for all patents; but utility doesn't generally require a working device (even for patents of a "device" type).

Whether it should is an interesting question, but if its clear from the information in the patent and the information generally available outside of it that a practitioner in the field could derive real utility from the described mechanism, I think there's at least a clear argument that actually having a demonstrated device is unnecessary, while with ones whose claims appear to violate physical laws or be of a class where claims have shown to be impossible to reproduce by other practitioners, the demonstration of a working device to demonstrate utility is important.