This is actually an interesting idea. Currently Node IDs are arbitrary ints, which makes it easy to have multiple clients node-wise. What matters is that the nodes have a unique ID, and that the total number of nodes is not exceeded.
If busses and buffers worked that way, and if Synths explicitly kept track of buffer and bus usage, it would a) be much easier to deal with them in multi-client situations, and b) we would have the makings of a way to very easily model dependancy. Then node order could be implicit rather than explicit, and parallelism could be inferred.