then i think we could implement complex State machines with multidimensional states.
Well, actually, just a small subset. Those whose response to an event may only depend on one dimension of the state. Let's call them linear multidimensional state-machines. Those are pretty uninteresting, and can be simulated easily by an aggregation of unidimensional state-machines.
In practice, you will find that there are interactions between the state dimensions and so, in order to model the responses to the different events that may array you will have to take into consideration several dimensions. In other words, behavior can be modeled as a multidimensional matrix of responses.
My experience is that when one gets to the point of needing a multi-dimensional non-linear state machines to solve some problem* it just means that he is doing something wrong and needs to get back and try to break the problem in simpler ones.
*) unless there is some kind of code generator doing it!