I did some rough benchmarks and I could hardly see any notable performance difference between the sorted and unsorted graph, even with very large numbers of SinOscs (up to 1000). Maybe someone can come up with an example where there is a significant difference?
Note: For now I only tested with scsynth which has a single array of wire buffers that is used by all Synths. In Supernova, however, every Synth has its own wire buffers (for thread-safety reasons), so the larger number of wire buffers could add up and harm overall performance.