I think you can have various ideas and thoughts about this. I have both types of "architectures". Mainly my older designs had more of the type "one for each" creating flows with a lot of nodes but in my later designs, I try to make "smarter" solutions using less nodes. It's a matter of personal evolution I think. What I am able to do today, I was not capable of earlier due to lacking know-how. A good example excercis of "node reduction" I experienced myself is here: Video device monitoring - typically wifi ip-cameras
Another opinion is about maintenance & readability. If you create a "very smart flow" it can at a first glance be difficult to fully understand the logic if you have to dive deeper into it in the future. On the other hand maintenance might be easier than if you have many nodes scattered all over the place that you need to modify for some reason
From performance point of view, I do not think there is a significant difference unless you have many function nodes in one of the others. But if you are talking about controlling lights, you have a RFXtrx, right, the most of the wasted time is consumed in that device anyway
