Two weeks ago I was in Rome, in the ITC. As you can guess from the title, I had a quite interesting, but byzantine, discussion. I was presenting a paper (with B. Iniguez, M. Shur and some other people) about a universal procedure to model TFT. The core idea of the procedure is using a compact model to model an intrinsic device, and then model all the rest of the parasitic elements as lumped elements. The other guy was telling me that this is not physical. Obviously, I don't agree. Moreover, I think that using lumped elements is near as physical as it can get.
My point is: it is OK trying to put everything inside your equations. However, this usually leads to a loss of insight on what is happening in the field-effect device. I think it is better to model the device with as much accuracy as possible and without losing the insight, and then add as many elements as necessary as lumped elements. Thus, the resulting model will have a quite straight physical representation, easy to grasp by intuition, and will not lose precision due to any approximation. Moreover, the simulator will handle quite effectively the model, because it has been programed to do so. Probably, it will even behave better, because the equations will be easier to handle.
Anyway, it is only my opinion, and there are some argements in the other side that I'll discuss another day.