I think the problem with dust is that they don't have any radius themselves, they're just points with a small image that scales up as you zoom out. I don't know if this is true, but that's my hypothesis of how they are.
So, since they're points, they can't collide. But if a point gets inside a 3-dimensional body, it could be registered. So particles could maybe have an "add mass" feature. You could set how much mass particles would add to real bodies they collide with, so you could use dust in protodiscs to feed existing "seeds" of small bodies. This isn't implemented, but I guess it could be. As long as particles are just points, it wouldn't be possible to make them form bodies.
...Unless you have a way of registering if they get very close to each other, which would count as a collission, and merge them into a small 3-dimensional body. I just think the problem with dust is that it'll take loads of CPU if they affect the rest of the simulation with gravity, or if the simulation has to check how close every single particle is to the other bodies every single time step.
This is just my understanding of how the program works. It may not be correct.