Overall this number is fairly meaningless on its own. It's like saying "my car is faster than yours because it has a bigger spoiler". That can be true, but there are almost always other factors involved.
The GPU in the 360 runs at 500mhz, and RSX in the PS3 is believed to run at 550mhz. 500mhz means 500 million cycles per second. Some very simple maths shows that the 360 can set up one triangle per clock cycle, and RSX takes two cycles.
Pretty much everything on a GPU can be measured by how many cycles it takes to process, but that's not the whole story either.
A GPU is not a CPU, it doesn't do one thing at once. It's a massively parallel machine with a large number of separate systems all working in a work pipeline. That is; the triangle set up engine will be working on a triangle, while the shader core will be shading completely unrelated pixels / vertices, the texture fetch unit will be reading various texture samples and the memory system will be reading memory from all over the place. This is obviously a massively simplified view of how the system works.
So if it takes 100 cycles to process a vertex (a typical vertex shader complexity), then that means there is a 100 cycle latency on the next step in the pipeline. Performance is not a 1+2=3 thing, it's a max(1,2) = 2 thing.
The point is that the overall GPU performance is only as fast as the biggest bottleneck. Each system will be utilised at a different level, and vary rarely is every system being utilised anywhere near 100% capacity. For example, there are few reasons for doing texture fetches while rendering a shadow map - but the triangle rate, depth testing, ROP rates etc are very important for shadow maps - because they are the bottlenecks.
Technically, RSX has more grunt than Xenos in the 360 - it can do a higher number of theoretical operations per clock cycle. However the architecture is quite different, the 360 importantly has a unified shader core (which means vertex program and fragment program processing and texture fetch units share the same hardware resources) whereas on RSX these are two distinct units (with their own limits and latencies). This type of design will typically mean that Xenos has higher average utilisation, and therefore greater efficiency. It also means that getting better performance out of RSX will be more difficult - as it harder to balance the workload for high utilisation.
This is partially why the SPUs are heavily used for vertex/triangle processing on the PS3 - as it moves the bottleneck off the GPU and onto the SPUs, allowing the rest of the GPU to achieve higher utilisation at the (often large) cost to CPU resources - I believe uncharted 2 spent around 40% of its SPU time on vertex processing.
I hope this helps explain why GPU performance can't be measured in one single statistic.
A much better analogy would have been engine displacement. That's basically what these polygon rates are telling you. It's an important thing, but not the singular most important thing.
