Hi,
I have had a long discussion with an Agilent expert and now I have the ideas a little more clear:
S-matrix depends only by the network and not by the source/load impedance.
Setting the input & output ports at 50 Ohms will return me a S-matrix normalized to 50 Ohms (S_50). Setting the input & output ports at 75 Ohms will return me a S-matrix normalized to 75 Ohms (S_75)... and so on...
I can pass from one to the other with some calculus and that all, I don't need to close my network on matched loads to run simulations.
I hope I have correctly understood! :|