Unlike the effect layers, nodes can have multiple inputs. Also, in more advanced node systems, nodes can have types other than "Image", e.g. number, boolean, etc., and node inputs can accept those types.
Well one example would be a node that outputs HSV as seperate (greyscale) outputs, you could then operate parallel on all these strands and merge them together in a combine node that has 3 seperate inputs for HSV and gives you the full color again.
Practically nodes make it easier to deal with more complex operations than layers (e.g. operations wherebyou have multiple source images and/or multiple target images, like e.g. in compositing, CGI etc)
One of the examples show a merge operation though, which involves a sequence and therefore would not be possible with just a stack of layers.