didn't come out as brilliant as hoped

This is taking ages! I've got the fastest EC2 instance as before, it's been at this stage for around 20mins...

Thanks for your help
I still haven't had time to sort out materials and stuff, but I've done some more angles. Click to enlarge.

Thank you
I will post whenever I make any progress, but that's pretty slow atm... I've just made some vray materials, and added coloured stripes to the room, but the noticeable change will be when I apply the materials to the balls. There are 4 materials, and they will be applied randomly.
It's quite overdone - once you find vray or any other renderer, a lot of people just make a room and fill it with shiny spheres. However I've noticed most people just spread them around the floor, I've never seen them floating randomly in 3d before.

Model: https://docs.google.com/open?id=0BzUvtdlY91uaZGtyTUNmZ1FaaGc
Render, 1728x3456, 360x180°:
[xxl-img:16qftfz6]http://i.imgur.com/cRUFp.jpg[/xxl-img:16qftfz6]

Here's the first preview

Just realised model didn't upload.
It's 156MB! where can I upload it?
Eye candy will take a while also, rendering will take ages!

I'm sorry, but I don't understand some of what you just said.
In my model, there are multiple instances of one component, "box", "box" should contain one group, a cube, which sketchyphysics considers a sphere. My plan is to run the simulation, then replace the group with a sphere, so I have a room full of spheres.
I get a bit scared with all these hacks, command-line things and BIOS settings, and I'm not sure what you mean by memory allocation, but when running the simulation, 2.8 out of 4 GB are used in total, and 1.6 out of the 2.8 is sketchup.
anyway I felt a bit guilty wasting your time on my little project, so I found a workaround. I restarted my computer, temporarily uninstalled all other plugins, ended all un-necessary processes. Ran it once until it crashed, to see how far I can go, which was 383 frames, then used the recording function to get to 380 precisely.
I can't remember distributions well, but I wanted relatively even spacing, but a random feel, I think that's "poisson"? you are right, they are a bit denser around the starting points, but I can live with that.
Here's the final result, with 1296-sided spheres. I will use vray to make some cliched eye-candy! The walls will be very colourful and stripy, and the spheres will be a mix of glass, frosted glass and metal, the ceiling will be a light-source and the ground a mirror. I will make a 360x180 panorama from where the camera is now.

Erm, my old machine is 64-bit, and the graphics driver is 54 bit if that's what you're talking about?
Model is attached. The cubes are big enough, and I've told sketchyphysics to consider them to be spheres, because I will eventually make them high-poly spheres - the cubes are my way of lowering polys.
The cubes are basically occupying the same space as each other, and when you click "go" the force themselves away from each other. I've tried putting them in 0G and 1G - 0G provided something similar to what I wanted when I tried it with 400 instances, but I needed more.
Perhaps this way of initially placing the cubes is the problem? But I did try it before by having the top open, and the cubes falling from above - it was fine at first but still crashed eventually.
I don't want to use an array because it needs to be "random".

I just tried re-running my simulation a few times to make observations.
When I run it from the beginning, it gets a few hundred frames in, before crashing. When I split it up into segments of around 75 frames, using the record feature, it can only last around 2 segments, and on the third it crashes immediately.
Memory usage stays at 2.7/4, processor at 55. I presumed 5% was other processes, and 50% was sketchup using only one core, but actually it seems like it's using each core equally.

When I ran your model, it took 5 seconds each time I zoomed or orbited. Even if I just press the "zoom extents" button, it took about 3 seconds to figure it out

sp.skp

That's a shame, thanks for the help anyway. Quick questions:
I've never heard of LARGE_ADDRESS_AWARE, is it enabled by default, or how else do I enable it?
Also any other ideas for this situation? It's just a bit of fun, but I want to fill a room randomly, but roughly evenly, with around 1152 spheres.