Improve performance with large number of onthefly objects?

[Fox5]

Hey all,

I'm currently working with a file that attempts to render 20000+ on the fly objects. (each about 12 polys) I'm currently just doing the onthefly drawing of the object once, storing it, and then copying it as needed. And performance is pretty bad.

The original development computer I was working on was a Pentium 4 with a Geforce FX 5200 and performance was hovering around 10fps. To test to see if it was the limited hardware causing the problem, I published an exe file (coming out to a ridiculous 3.05MB) and attempted to run it on a computer with a Radeon X1950XT and a Core 2 Duo. Performance was actually slightly worse, which leads me to believe that either (and perhaps both):
20,000+ objects is just too cpu intensive (no idea how I would reduce the load)
A 3.05MB exe file for something that only contains a few hundred lines of scripts does terrible things to the cache of a cpu

Either way, it would appear that the polygon budget isn't a limitation given that performance actually went down. So, given that, is there anything I can do to reduce the cpu load?
In fact, I'm sure it's cpu, as my cpu load spikes to max (on a single core) when running the app. My memory usage also spikes about 300MB.

Also, is it possible to add both onthefly objects (which stack with .startlayer) and static models (which stack with .addchild) to the same node and have them translate and scale together?

Thank you in advance.