The current "graph size" includes deterministic nodes and thus it doesn't really signal the model's complexity. I sometimes miss the opportunity to see how many parameters are estimated (i.e. sampled or stochastic) relative to the number of data points that go into the model. Right now, for example, I'm trying to see how increasing the size of dmnorm() affects graph size, but I'm unsure how much is deterministic (i.e. not really an increase in model complexity) and how much is stochastic. So something like a "stochastic graph size" would be cool. Certainly on the "nice to have" list rather than the "need to have".
This is now in the development version (future JAGS 4.0.0). When the model is compiled it reports the number of observed and unobserved stochastic nodes, as well as the total graph size.