The manual pretty clearly states "If different objects use different values, then the Pathfinding behavior must generate multiple obstacle grids in memory, and pathfind along them separately."
For starters, the cell border size (what I'm assuming you're talking about regarding padding), does not have any affect on the actual pathfinding grid itself at all. So if you take that out of the equation, it should be clear enough that a 32 grid size is not going to be the the same as a 31 grid size.
As for which grid used when finding a path for a particular object, its simply one that is defined by the pathfinding behavior properties that are attached to that same object.
While I'm not sure if rewriting the manual entry would help someone understand these things any more than it already does, I do think an in editor visualization sure would help a lot.