BTW, no amount of tweaking of firmware is going to make Z-Wave a high speed network. So, to the degree that low latency feedback is important, and it often is in order to be useful, a multi-hundred module system is probably going to be a bit over the top for Z-Wave. If the average poll time in even a small network (where the controller can get to all module directly and not require extra routing) is in the 1 to 2 second'ish range, even if you do overlapped polling (which we do), 200 modules is going to have pretty bad latency if all of those need to be two way.
If you keep 5 overlapping polls going at any one time, and all of the modules are completely happy and respond in the 1 to 2 seconds range, you'd take 40 to 80 seconds to make it around the list and get back to the start. If you tried to keep 10 going at once, it would be 20 to 40 seconds. And the bigger the network the more hops any one message will have to take along the way, which also slows the response time, so the above numbers would actually probably larger.
I'm not sure what the limit is on how hard you push the number of overlapping polls, though we'll be exploring that, but at some point you will overwhelm the Z-Wave network (which isn't very fast) and have so much simulataneous activing going on that it will just bog down.
If you treat many of them as just one way, it would obviously be a lot more practical. But how really desirable is it to have a large building of things being controlled without feedback to the control point about what those things are doing? Or to have that feedback be delayed by a minute or more?
To me, I think that there comes a point where you'd want to use something a little more upscale as you crank up the number of things under control, ultimately moving to a wired system.