Traditionally the criteria for the bittness of a cpu, is its address length. Given that, there's hardly any need to use 64bit CPUs in Consoles for the next generation. The annoted performance gains in UT2003 can be attributed to the larger amount of registers and other isa specific enhancements of x86-64. The EE for example implements a subset of the 64Bit MIPS IV ISA with 32 Bit adressing, yet its word length is 32 Bit, so this processor would not qualify as a 64bit (every Pentium also has larger datapaths, FPU, MMX, SSE, etc..) cpu in the workstation realm. Point is however that on the computational side 64 Bit (with all processing done on 64 bit words) processors are only at an advantage when the 32Bit domain offers unsatisfiying precision for a significant share of the computations (and thus higher precision can only be archieved by multiple instructions). If 32 Bit precision is enough you profit from (potentially faster) smaller hardware implementations of alus and most important a much lower data/code footprint. To sum it all up, generally higher bittness hardly equals better/faster and 32Bit computining will IMO remain the sweet spot in embedded markets for years to come (until 4Gb adressing will become a necessety there).
