Interesting idea. Although, I'm not sure the results will be different compared to if you just allow the magnetization to be free. ATK will adjust the magnetization to its minimum energy configuration for each lattice constant, so in the end you should end up in the same global minimum (provided it's not blocked by very high energy barriers separating local minima, but that's a risk you always face with any kind of optimization).
In the end, the magnetization is just another degree of freedom, so the situation would be quite similar if you try to optimize, say, the a and c lattice constants of a hexagonal structure by keeping a fixed (lattice constant) and scanning c (magnetization), or keeping c fixed and scanning a. Running the self-consistent loop is pretty much the same as scanning the magnetization.
