BOINC relies on redundant computing both for validating the science, and for guarding against cheaters claiming too much credit. As long as a majority of users isn't cheating, this system works.
But, Rosetta@home isn't currently using any redundancy, meaning only relies on anyone crunching incorrectly gets "impossible" result or something, while for credit there isn't really any limits currently...
BTW, even without anyone cheating, the BOINC-benchmark can give wastly different claimed credit on different computers.
Rosetta@home is currently working out which method to use to minimize the variations in claimed credit, and to guard against cheaters.
Optimized BOINC clients was made so someone running an optimized seti-application should get roughly the same claimed credit as other users. Running an optimized BOINC client in other projects on the other hand means you're getting artificially high benchmark-score, and therefore also artificially high claimed credit. In other projects this isn't really a problem, since due to redundancy the too-high claim is discarded.
So, till Rosetta@home has landed on a solution, using an optimized BOINC client can be seen by some as cheating...