There is no such rule that the dominant pole can't be at the output.
I think you meant cascode, not cascade structure in your post.
It is not about being cascoded or not. It is more about being 1 stage amplifier vs. two stage. You remember that all we want to have for a stable negative feedback amplifier is the loop gain crossing 0dB with a -20db/dec i.e. like a 1 pole response. In a 1 stage cascoded amplifier which has at least 2 poles, the non-dominant pole is at the cascode transistor source and is usually at some fraction of ft of the technology. Since it is at such relatively high frequency and is more or less fixed, it is very easy to compensate the loop by just having enough load capacitance at the output and thus making the loop gain -20db/dec at the cross-over frequency.
In two stage amplifiers situation is a bit different. There the two poles of the open loop amplifier (one at the 1st stage output and another one at the second stage output) could be quite comparable in value. In this case it is better to use Miller compensation because it just does what we want - pole splitting. It pushes the dominant pole at the output of the 1st stage towards low frequencies and pushes the non-dominant pole at the output of the second stage towards high frequencies.
Now, if in the case of a two stage amplifier you drive a very low load capacitance, so that the 2nd stage pole is at high frequencies, then you can still use load compensation, connecting more cap at the output of the 1st stage and thus making the loop gain cross 0dB with 20db/dec. Or vice versa - if for some reason the 1st stage loading is very small, then you can connect more cap at the output of the 2nd stage and still have -20dB/dec crossing. But in the general case this are not optimum solutions for a two stage amplifiers.