Hmm, this is kind of an interesting idea, but I share others concern that it is unlikely to result in setups that are any better than a manually created setup.
However, I think there could be some interesting research done on the dataset that you gather, but simply averaging values probably wont reveal anything very interesting.
What might be interesting however, is to produce some statistical data on certain parameters and extrapolated behaviours. For example, we can fairly easily determine things like basic spring balance, aero balance, critical damping percentages, top speed, etc, from the setup data. A statistical analysis of the range of these parameters, cross referenced against track say, could be pretty interesting.
Also, you could plot one setting against another setting, to determine things like how much more basic spring oversteer do you need to add when using a locked diff ? Or what is the average critical damping percentages used at different tracks to determine people's response to different bumpiness profiles ?
I would suggest dumping all the data into a database and running some more complex queries on it.