In case anyone else is reading this, and it’s helpful, here’s some additional corroborating evidence that there exists a mode where the VRCS2 loads are included, but the original “factory default” button->load association persists: here are the instructions for getting this behavior on a Vera: http://forum.micasaverde.com/index.php?topic=23497.0
The Vera also knows about scene controllers (SmartThings does not), so there is another method for getting things to work where you associate the scene controller buttons with the loads via the Vera controller (http://forum.micasaverde.com/index.php/topic,13251.0.html). The reason that even Vera people want the direct association is that it is faster to respond and works even if the Vera controller flakes out. For SmartThings, at least with the current level of software support, preserving the “factory” association would be the only way to get the buttons to do anything useful, and it sounds like @jensed did manage to get it to work this way… somehow.
It is unclear if SmartThings has the right degrees of freedom to accomplish this inclusion process in a deterministic way (even on Vera it appears to be partly stochastic). From the Vera forums, it seems that the critical thing is to not let the Vera “configure” the controller portion of the device. Interestingly, SmartThings shows an option to “tap to configure” the controller after you include it, and its possible to skip this step (at which point it shows up in a list of “Yet to Configure” devices). Unfortunately, for me at least, skipping this “configuration” step didn’t change anything, but I did go so far as to cut the circuit breaker (as in the Vera instructions) yet.
In general, I wish there was greater clarity about what happens during a “configuration” (and also, exactly what state changes during an exclusion, inclusion, etc.). I get the impression that the whole stack is full of leaky abstractions and weird corner cases where past configurations persist in spite of resets, exclusions, etc. I feel like I don’t have an accurate internal model of the Z-Wave stack that enables me to understand why tripping breakers at those stages would make a difference. If anyone knows about decent Z-Wave documentation, I’d love to understand it better.