Which is better? It depends on what the purpose is. If you don't know, RISC stands for Reduced Instruction Set Computing, and CISC stands for Complex Instruction Set Computing. Thus the performance depends on the type of program. I heard that CISC employs tons of instructions which use higher level programming language concepts. I just looked at the Intel IA-32 programming reference manual, and found several such instructions:
ENTER - makes a stack frame
CALL- store return information to the stack, and goes to a procedure - it could be an absolute memory address
INTERRUPT - interrupt return
JUMP - jump to a different part of the code
LOOP - loops according to the EAX register count
MOVS, and others - mode data from a string to string
POP, POPA/POPAD - pops data from the top of the stack
PUSH - push a 16, 32, or 64-bit value onto the stack
RET - return, return from a procedure call
These instructions all are virtually small programs. None are supported by RISC architectures, such as the ones found on all consoles except for the original Xbox. The end result is that the programs are far smaller. It makes a difference. A running program which would be say, 100 MB on the x86 CISC architecture is 300 - 400 MB on the RISC one. This is due to the far higher amount of instructions required to perform similar commands. Thus - saving on RAM usage - which was the primary reason that the CISC architecture was originally adopted in the 70s.
The benefit from running RISC is that far fewer transistors are used in the instruction stage - allow for small chips, and more SIMD performance in relation to the rest of the performance. However - the CISC architecture is far faster at programs, as this microcode dramatically speeds up the rate at which programs are executes. That makes the critical difference, and is the reason PCs only use x86 architectures.
Thus it's a question as to which is more important - overall program performance and memory utilization, or SIMD program performance. Thus I prefer the CISC architecture, since it allows for the game program to execute faster . As for SIMD performance - I leave that up to a vector co-processor, which is RISC in design. Such as the GPU, or even better, the on-chip GPU.