GH) > then define (still in GH) some instance definition (or many: case variants) > then place it according some "policy" (3d point grid and the likes). Note: Only doable with code, mind (C# in my case).
Obviously you can skip the creation part and instruct GH to deal with instance definitions already listed in the Block Manager (say: find the block named "cell666_B3" blah, blah) ... but that means that you can only use them (meaning a rather "limited" parametric approach) and not make them from scratch (meaning a true parametric approach).
But I guess that you've tried the block way in the Rhino environment already. That said I use rather solely this approach in GH and yields quite manageable object collections - I would say "real-time" response (up to 20K instances) but I use dedicated Xeon E5 1630 V3 workstations (with NVida Quadros K4200 and up for the graphic response part of the equation) so the "performance" is rather a subjective thing.
Modifications:
easily doable with GH (on instance definitions at placing time: since you need only to scale them and not vary their topology).
Anyway post a portion of the R file.…
FORE MeshMachine (rather better) or after
BTW: For a mesh with 7M points ... well... you'll need some proper CPU to deal in a reasonable amount of time (what about a Xeon E5 1630 V3?).
Alternatively find a friend who knows very well Modo ... and see first hand what the US Movie Industry is all about.…