It’s usually reasonable to group objects into layers according to their purpose in the scene, and to cause a particular camera only to “see” those layers that you know it should consider. This will save computer-time, as will breaking assets into separate libraries and linking only to those that you need. It requires careful advance planning, and consistency. You need to come up with a plan, write it down, and use it carefully for all shots and all components thereof. (Every shot has its own “breakdown.”)
Usually, I set up a master file with all of the cameras, and links to all the objects, and from these I have a “shot file” which links to a single camera and to the visible assets. (Thus, two levels of library indirection.) The “shot files” use “scenes” to provide further breakdown, and all of these are eventually rendered. It’s a lot of paperwork but it saves hours of time … and time is what you just don’t have.