I think we may need a 4th gen BOINC credit mechanism or some form of benchmarking on an individual project basis (without the need for 2 weeks of crunching) to estimate potential RAC.
I wrote the following in the 'for the betterment of BOINC' thread:
If we were able to benchmark attainable RAC with the user's computer (through a project specific benchmark perhaps) then they would be able to work out how much GRC they could earn ahead of crunching (currently you need to crunch for weeks before figuring this out), this could help users make a more educated project selection.
Likewise, on the topic of RAC - there was a large thread on a possible '4th gen credit mechanism' which attracted a lot of attention last year: https://boinc.berkeley.edu/dev/forum_thread.php?id=10953 It would be excellent if further research/discussion was held on an improved credit mechanism.
well actually this seems to me more a problem of number crunching out of already existing data sources. For sure this will be the hell of a work and in the end you would end up just with an average per CPU/GPU combination.