Man but do these guys get it wrong. First of all, no doubt Sony is pulling a hype trick - but that has nothing to do with those benchmarks. What's even worse is that that site insinuates IBM was testing PS3 rather than just Cell itself. As if IBM cares about PS3's performance.
The numbers come from these tests here from last year:
http://www-128.ibm.com/developerworks/power/library/pa-cellperf/
There was extensive discussion on the subject back in the day... but what's even
worse about that article is that they take the real world Flops performance of Cell and compare it to the theoretical* performance of the XeCPU. I hate to bust up the party, but not only is the real max theoretical Flops count of the XeCPU more realistically ~75 GFlops, but it would never reach that number itself in a real world situation either.
* No CPU can or will ever hit it's theoretical maximum under practical conditions.