So let me level set before we try finding a solution.
You can actually construct a frequency multiplier by passing a signal through an XOR gate while the other input is tied to the delayed version of the first input. For example you can tie input A to the signal and input B to the signal that passes, lets say, 10 invertors. Now this, because of the propagation delay, will generate pulses where the two inputs were out of phase with each other. And since this occurs twice every cycle, we have an output thats twice the frequency of the input signal.
The question now is on a similar setup or a complex one is it possible to come up with a frequency divider. Your multiplexor example is still not good, because you will tie the output to one of the inputs of the multiplexor thus a feedback from output to the input exists.
More bluntly put, is it required to "remember any previous value" for a circuit that you would construe to divide the frequency?
I hope I made it clear. It may happen that we do not have a circuit at all. But if we have that'd be a great debate on it!