I don't know how they measure,....got to get the parallel action going...will require some real programming.
As far as multi-threaded performance goes... IBM's compiler supposedly reconfigures single-threaded code into multi-threaded code. I don't know how well it works at this point, but it'd be quite the boon to development.
Regarding Cell yields... it's The Inquirer. Believing their non-sourced stories is like believing The Weekly World News.