The extreme of game/engine split, break minetest down to a core functionality of maintaining voxel areas and providing an interface for manipulating them, real-time mesh generation. Offload everything else to whatever other program (ie. game engine) wants to utilize it. Most games/mods, as they exist, would probably even be maintained (moreso those concerned with block definitions).
Would it be valuable? Would it be valuable to minetest_game? Reading the outstanding issues, reading this forum, it seems like a definite 'maybe'. The muddiness of the line between engine and game seems like the core problem of the project.
If nothing else I feel 'tidy up game/engine split' should definitely be added to the roadmap. Basic stuff like splitting up the config file, devolving appropriate settings to the game or isolating them to the server.