I don’t completely disagree with you, but I would point out the paper ‘A Higher-Order Graph-Based Intermediate Representation’ by Roland LeBia, et al from the U of Saarland Compiler Lab. That paper, which describes the groups Thorin IR, does show some advantage to the sea of nodes based approach (although the representation may not be classified as purely SoN). The advantages seem particularly compelling for mixed functional/imperative source languages, which was essentially their goal.