Dear Stephane,
thanks a lot for the clarification!
For the 3rd order approximation, it seems that Ian has posted code to do the simulation manually using the stored decision rule (though dynare_simul_ probably does the same) [Manipulating the decision rules).
Best,
Thorsten