I am trying to run a pd set which includes many abstractions and subpatches. In order to manage space, I have used many many different [s~ $0-toSomewhere] objects in subpatches, with the [r~ $0-to Somewhere] object being in another subpatch.
Does this overload or just load a little bit more the cpu?

I f I have to choose for the less to max cpu consumption, what is the best practices:

1. Use straight lines connecting inlets and outlets and abstructions in the set.
2. Use [s~ $0-..] & [r~ $0..] objects in the main "page" where all abstractions are.
3. Use [s~ $0-..] & [r~ $0..] objects in subpatches, where the object is in different subpatch from the [r] object.

Or
...It doesn't matter the cpu how these connections happening (or little difference between the choices)...?

from my experience, (1) is the best option for saving cpu.

if you have a patch with lots of s~/r~ or throw~/catch~ pairs, the cpu does add up a little bit.