1. Optimization often has a critical effect, but in some cases with very regular structures, like graphene or nanotubes, it may be smaller, and you can get away with e.g. moving the nanotube rigidly instead. So it's not always necessary to relax the whole device, but it is always important to make sure the setup is reasonable. In the case of a molecule, this could mean relaxing the molecule itself, and then optimizing the distance between the surface and the molecule, but ignoring (to first approximation) and influence of the metal on the internal arrangement of the molecule, and conversely assuming that the surface remains intact even when the molecule docks it. All this needs to be clearly stated when the results are published, and can be subject to criticism by a reviewer, and so any approximations made must be justified in the article (e.g., "we tried two ways to positioning the molecule on the surface which gave similar results, so it was assumed that the primary effect was caused by the distance from the surface, not the molecular configuration itself").
2. Indeed, as mentioned just above, the electrode has a critical length. If you are below it, you get bad results, if you are above it you get correct results, but no matter how long you make it, you still get the same result. For details, see
http://quantumwise.com/publications/tutorials/mini-tutorials/99.
