steinandre, you may better use ANSYS in such a case, because DesignSpace is designed, to my know, only to help the designer have a common feeling about s structure, so it is not too accurate.
To your fisrt question, i think, it is a general case, that the FEM-Models are stiffner then the real models (because of discretitations), i mean, that this is not only the tetrahedron case (think about shear locking phenomen). Normally, the higher the polinomial order, the less stiffner the model is. So the TET10 (parabolic) will approximate the displacement better then the TET4 (linear) and so does Quad9 to Quad4 (Shell). But, the calculation becomes more expensive also. Any comment?
cheers