I didn’t mean to post my comment above as a reply to Ben Kelly. It was supposed to be general. My bad.

There is so much to say in this subject and I would say that it’s a common misconception that the PS3 is more powerful. In certain aspects, sure it is. Looking at extras like bluray, wifi, gigabit and such, sure it has better hardware. But, if you’re looking at performance for different workloads, I would say that the 360 is faster in a lot of scenarios.

It’s sad that fans just think about the CPU, not about ram or GPU, these two are as responsible for good performance as the CPU. Let’s start talking about the CPU.

CPU: The Cell is faster for most work, especially work that can be paralleled. But I wouldn’t be that sure that the Cell is faster for traditional CPU use, since its SPE’s are more suited for high throughput parallel work than general purpose computing.
Think of it this way, Cell has one real general CPU, while the 360 has three, on the other hand the Cell has 7 SPEs, which are very powerful, but only for certain tasks. So in a good scenario for the PS3 the Cell will kick the 360′s ass, big time. On the other hand, in a scenario not well suited for SPE’s the 360′s three more general CPU cores may as well be faster.

RAM: PS3 has 256MB ram and 256MB vram. The 360 has 512MB ram that can be used as the developer wants. So using 400MB as ram and 112MB for textures is not a problem.

How’s the bandwidth then? The PS3 has 25.6GB/s for its RAM memory and 22.4GB/s for the VRAM.
The 360 has 25.6/GB/s to it’s combined memory when the CPU is reading and 22.4GB/s when the GPU is reading data. The 360 GPU also has 10MB of edram with a bandwidth of 32GB/s.

So, in a scenario when a developer needs no more than 256MB ram and 256MB of vram, the PS3 has higher bandwidth, since the GPU and CPU doesn’t share memory. On the other hand, as soon as a developer needs more ram or vram, he needs to start borrowing and doing transfers between the two areas, turning the dedicated memories into a performance hog.

GPU: The GPU in the 360 is more powerful, hands down. It is a newer generation of GPU than the one that can be found in the PS3. In the PC world the 360 GPU would be called DX10 class GPU while the PS3 is a DX9 class. The 360 has programmable shaders while the PS3 has non programmable shaders in the form of pixel shaders and vertex shaders.

So what does the programmable shaders do? They can do the job of the different unprogrammable shaders the RSX has and it can also do general purpose work instead of the CPU.
The PS3 has 24 pixel shaders and 8 vertex shaders, while the 360 has 48 programmable shaders. Now the 360′s GPU is more powerful in numbers, but it’s also more adaptable to workload, since it can behave like 48 pixel shaders if that’s needed or 48 vertex shaders. The in a scenario where few pixel shaders and lot of vertex shaders are needed the PS3 will have its unprogrammable pixel shaders doing nothing while the 8 vertex shaders are undermanned.

The edram used in the 360 gpu also makes anti aliasing much, much cheaper performance wise.
http://en.wikipedia.org/wiki/RSX_%27Reality_Synthesizer%27

Let’s sum this up then:
If a PS3 developer needs no more than 256MB of RAM and 256MB of VRAM. He does not utilize anti aliasing and have a graphical workload that suits the split of 24/8 pixel / vertex shaders. He then sets up the Cell CPU so it can aid the RSX GPU in certain tasks. Add to this a highly parallel workload that can harness the Cell CPU. Then the PS3 is faster.

On the other hand, consider a scenario where anti aliasing is used, the graphic changes its demand of pixel / vertex shaders depending on area and the game shifts its need between large textures and large ram usage. The CPU requirements are mainly traditional workloads. Now the 360 will be a lot of faster.

The thing is that it’s a lot of easier finding a scenario where the 360 is fast and most multiplatform games will certainly benefit by the flexibility in the RAM and GPU of the 360.