Thanks, that clarifies the reason for the limitation. To be honest, we kind of suspected that was the reason, but we weren’t sure. 
For now, our plan would probably be to store the procedural definition of player-created objects (roads, buildings, etc.) in our own save format, and recreate the corresponding virtual models when loading a game. We are also considering storing the generated mesh data as a cache, so that complex geometry would not necessarily have to be regenerated every time a game is loaded.
In that case, the procedural data would remain the authoritative representation, while the cached mesh data would only be an optimization and could be regenerated when necessary. This seems like a good fit for the custom-files/virtual-model approach you suggested. If you see any drawbacks with this approach in Flax, or if there is a more idiomatic way to handle this kind of runtime-generated content, we would be very interested to hear your thoughts. 
Before answering your question about other APIs that could be useful to us at runtime, I would also like to understand one related part of the pipeline.
Does Flax already have an internal API for generating/simplifying LODs from a Model or Mesh, for example the same system used when importing model assets?
Model::SetupLODs() seems to let us define the LOD structure, but as far as I understand it does not generate the lower-detail meshes themselves.
If such a LOD generation API exists, is it currently editor-only as well?
For procedurally generated models at runtime, such as roads or buildings, having access to both model persistence and LOD generation could be very useful!
More generally, I was also wondering whether it could make sense in the future to expose some of the asset creation/saving APIs to runtime builds on platforms where writing game data is possible, rather than making them strictly editor-only. I realize this may be much easier said than done, especially if parts of the asset pipeline depend on editor-only tooling or libraries, but it could be useful for games where players can create persistent content at runtime.
Besides procedural content, I could imagine this being useful for modding as well. For example, a game could potentially import or generate new assets after it has been shipped, whether they come from player-created content, mods, or additional content distributed by the developers themselves. I’m not sure how well this fits with Flax’s asset/content pipeline, though, so I would be interested to know your thoughts on whether exposing some of these APIs selectively on platforms that support writable game data would be practical.