The main point not supporting this feature is that, in our opinion, there are not many applications in graph visualization which really require hyper edges. Therefore, we would like to hear about your use case.
From a conceptual point of view, it is often possible to model a hyper edge with a complete subgraph in the traditional graph model (each node is connected to each other node). Since this transformation typically results in a large number of edges, computations on hyper graphs are often more expensive than for normal graphs. And as additional disadvantage, since a larger number of edges results in general in a less readable diagram, these visualizations are often not very readable.