In general it’s not going to work to have any type of VT used from a RVT. This means you can’t use streaming VTs to generate a RVT, and you can’t use one RVT to generate another RVT.
The problem is, a RVT is basically a cache for the material results. When a material is using a VT, the VT will not necessarily have the final data right away; it will either be streamed in or generated across multiple frames. You don’t want the RVT to cache the material before the dependent VTs are ready.
To make this work properly, we’d need some way for the material to signal that it’s using VTs that aren’t fully loaded/generated yet, and trigger the RVT to continue updating the material until it’s ready. This adds complexity, and it also reduces the efficiency of the RVT, since it’s no longer enough to simply cache the data once. Maybe this will be solved at some point, but I don’t think it’s something we’re planning at the moment.