This, indeed, is a question which is not easy to answer.
1.) What kind of result do you expect from the analysis ? Only stable/unstable ?
Or do want to have a stability margin (magnitude or phase or a combination of both ?)
2.) In principle, it is quite clear: If you have three different loops, there will be three different loop gains and , hence, three different margins. And than the question arises: Is there any loop resp. margin which governs the system behaviour ?
In many cases the answer will be "yes", but one cannot give a general answer, because it depends on the loop arrangement.
3.) To avoid some errors: It makes no sense to reduce a four-loop system to a single loop since there will be exactly four different ways to do this - all with different results (thats perhaps surprising, but it is true !). Of course, in each case the closed loop transfer function will be the same (but not the loop gain function).
4.) Fazit: You will have more than one expression for loop gain and more than one margin; and than you have the choice to interpret this result.
That is the problem.
LvW