My experiments with using the Roland CNC SRP didn't turn out too well. Although the GCode program worked well, it turned out to be very difficult to get consistent heights for the parts. Out of every run of eight parts, I would have to throw away three on average for being too short. I suspect that this was because the end mill exerts an upward force on the parts as it cuts. Normally this isn't a problem, but if you have eight slabs of acrylic in a vise, the acrylic-on-acrylic friction may not be enough to hold it in place no matter how much you dare tighten the vice.
I did manage to get the GCode program down to an hour, but the other problem was drilling the holes. The drill bit kept heating up and melting the plastic, and I had to flood the coolant onto the part and bit to get it cool enough. The Roland doesn't have flood cooling, so I ended up slathering coolant constantly onto the bit. And even then, sometimes the heat was just too much.
So in the end, my hypothetical cost to produce a single module went down only to 56% of the original contracted metal parts price instead of the 39% I originally estimated.
But, I had a new idea where I might be able to cut an entire module side at once using the Trotec laser cutter. My estimate for the cost using this method is back to 39%.
Part of the method involves cutting 1/8"-deep slots, and I figured out a way to do this with the laser. I just engrave at various speeds until I find the speed that cuts a slot as deep as I need.
Next week I should be able to cut the sides for an entire module, and then we'll see how it goes together.