One of the many reasons why its harder to port PS3 -> PC is because the PS3 CPU and GPU work much closer together, and that arrangement doesn't work on the PC architecture. The PC architecture lacks the floating point performance on the CPU to run the same functions there and thus much of the code to run the graphics subsystem has to be rewritten for a start. Furthermore there is far less bandwidth and far more latency between the CPU/GPU on the PC architecture than on the PS3. The Xbox 360 still does some calls to the CPU for rendering as well, but it isn't nearly as significant and it can be changed quite a lot easier. Farcry 2 from the PC to 360 used 2-3 programmers, to the PS3 it used 14 programmers to illistrate the difference.
If a developer doesn't want to push either system heavily, theres still no incentive to go PS3 -> 360 when the Xbox 360 SDK is far easier to use. If the developer wants to push both systems heavily theres still a requirement to nail down the memory management on the CPU and use the VMX units which both translate to good performance on the PS3 as well. SPE -> VMX or VMX -> SPE it really doesn't matter which way it goes AFAIK.
From what I can tell, a hybrid approach is also very popular. For example, leading on the 360 with most code and giving the PS3 seperate attention where it counts such using the RSX and SPEs together to give relatively equal performance.
Tease.
