At the moment, magic does not correctly extract a deep nwell structure with multiple isolated pwells in it. I have a development branch of magic that does that but it's still being debugged.

Will extracted isolated NMOS devices have the same model as the normal ones? If that is the case, I could simulate NMOS body-biasing with the current models.

Yes, the nMOS devices over deep nwell use the same model as nMOS devices in the p-substrate. I find it hard to believe that it really has no effect on the device model, but I have seen that for other foundry processes, too.