If collision detect were hardware (GPU) based, then I would much prefer it to be on a polygon level (i.e. each polygon of the collision body, for example a character's body mesh, is considered in the detection and not just simplified shapes,
Thats is essentially the way it works in nearly every other app
even Daz and poser and of course Maya’s Ncloth system.
I think RL has taken the approach of making it easier to export Characters &motion to other programs for more advanced physics, rendering etc instead of trying to compete with the other apps on a performance level.
those who remain exclusivity with the Reallusion ecosystem will have to accept it's built-in limitations
real-time performance (outside of gaming) is not the be-all and end-all.
“Realtime” animation has alway been essentially Iclone’s Major Selling point
and to a degree Iclone works somewhat like a realtime game engine
however when compared to other modern realtime engines ,like UE5, Iclone is a bit behind in performance ,particularly when you involve something other than just a character.