Would it be too confusing for me to say either approach is appropriate?
You can start your artificial boundary conditions either way - setting either forces or displacements. But whatever you use to start, you should take these results and apply the opposite (displacements or forces) to the missing systems and see what happens to them. As the other (missing) systems load up or move, you now have a new set of loads or displacements to apply to your original system. You keep on iterating back and forth across these artificial boundarys until they get back to equilibrium.
That's why your probably better off by modeling these systems together - the entire stiffness matrix will keep this "boundary" in equilibrium automatically.
With an eye on system stiffness, you can reduce model size. Say, for example, you have a 4 inch line off a 16 inch header. In terms of load and displacement, the 4 inch line will go wherever the 16 inch line wants because the 4 inch line cannot provide enough load to move the 16 inch line elsewhere. You can specify the 16 inch displacements as the boundary condition on the 4 inch line analysis and believe the numbers without iteration. That's what you do now for hot equipment connections. (Aside, but this would not work if you had 100 4 inch connections off that 16 inch line.)
Your piping system communicates forces and displacements through system stiffness. You cannot arbitrarily stop this movement of information (those forces and displacements) with your casual location of system boundaries.
_________________________
Dave Diehl