Right - Load and Append only during generating the cache (where you don't have to have more than 1 pass, and can do it every 5 or 10 or 30 frames, depending on how much things are changing). Then just Load during rendering.
And yep, accounts for all camera positions. Basically calculates it and saves it, loads what it saved, calculates it from the new position and adds that in to what's there already and saves that, loads that and.... and so on :)
You can test it with a simple scene, e.g. a cube where camera circles around it. Just calculate and save while looking at the front face, when animation moves to the back face, there's no Cache calculated for it. With the Load and Append generate process, then Load during rendering, there's info in the Cache for the front face and for the back face (the Cache now has info for various camera positions stored in it).