They're going to need a lot more proof before I'll accept it. Translating from one instruction set to another is not a simple thing. Translating system calls to equivalent system calls isn't trivial, but has been done. Translating between graphic routines is something I know nothing about, but something tells me it's not an easily solved problem. (3D cordinates represented differently? All it'll take is a matrix multiply to transform them... that's only 9 multiplies and 6 additions per cordinate)

VMWare does true hardware virtualization(translating system calls), and they do it well and fast. The only reason they can do it at any speed is because they run the native code and only intercept system calls.

If this were an easily solved problem someone would have done it. Maybe they've done everything they claim, but the likelyhood that it's efficient to the point of being able to run todays x86 apps on todays PPC processors is unlikely.

Matthew