As I mentioned previously, depending on the scientific question you are asking, you choose the right technique. So you will try to crystallise a protein to know its structure, but you might prefer NMR if you are interested in a temperature-dependent property, for example; or you might decide to use molecular dynamics (MD) simulations if you are interested in looking at the dynamic behaviour of your system.
In MD, sometimes this is not enough. Imagine you want to simulate TolC (outer membrane protein of the three component efflux system in E.coli.), in its physiological environment. This, using explicit solvent, would mean nearly 13000 atoms of the protein, around 8000 for the lipidic membrane, and probably 120000 atoms between water and counter ions, for a total of more than 140000 atoms…for people who have no idea about limitations in MD, this numbers are huge and therefore not easy to handle, especially if you want to simulate the system for at least some tens of nanoseconds, in order to reach an appropriate timescale. Still, you can do it, if you have time, good computers, and a keen Ph.D. student! More interestingly, you might concentrate on a specific topic and reduce these huge systems to smaller ones. You might look at single helices extracted from membrane transporters embedded in bilayers, or you can simulate only regions of the protein. So far, simulations of the three components assembly are impossible, at least at atomistic level. Probably the new coarse-grain techniques will allow us to go further these limitations, still giving biologically relevant results.