1) I doubt 50% was a scientific measurement, but a ball-park figure. 4 SPU's is hald the cores in Cell, hence 'half the power'.
2) How do you come to 70% figure after graphics, and OS work would take up 20%? The article sayeth 50% after graphics is left for non graphics stuff, and you seem to think that 70% of processing power after graphics should be available for non-graphics stuff. I can't see how you get to that figure.
There's no structured division of resources. One game might use 20% for graphics and 80% physics + stuff. Another might throw most of the processor into graphics, with less complex AI and physics (eg. flight sim) in which case 80% might be used on graphics and 20% non-graphics. This EA statement is only giving an example of one game. In this case they have produced their graphics and can't do any more graphics work because the GPU has reached it's fillrate limit. 4 cores are still unused at this point so they can be used for other stuff. That's it. There's no system-wide performance considerations that can be rationally gleaned from this. We've no idea how efficiently they're using fill-rate or how they're using the Cell for graphics. Could well be another game uses 80% Cell for graphics without ever hitting the fillrate limit. And there's no reason to think if OS work were running in the background it'd consume 20% of resources. And there's no evidence what sort of OS/non-gaming functionality might be running either.