I try to get clear ideas after reading all arguments posted so far in various forums...
What is interesting is to obtain a very large base of identical hardware that can be targetted by a developper who wants to make the effort to optimize his code in order to produce something great (if the number of potential users is low, it's not really worth the effort).
This optimization will be really great if it's done in order to take full advantage of a 'beast'. Recent GPUs (xbox360, ps3, ATI, NVidia) are pure 'beasts'.
In XNA forum it is said you can have 6 threads over the 3 CPU cores, but I don't think it will bring more reward than a true optimization of shaders in order to make GPU do all the hard work. But XNA point was to make things easier. Trying to make things done by GPU shaders draws that back away. It will be complex to learn and master.
About PS3, the lack of access to RSX is quite a problem... No real opportunity to tune something great. So, I think PS3 is less interesting than XNA for now. It's too close to a mere PC without any specificity. But, in fact, it's a good choice for a beginner that want to stick to classics (C language, unix environment) and don't want to be annoyed by hardware-specific programming (but in that case a mere PC could eventually do the same, so it would be, then, just the pleasure to have a 2 in 1 machine : console+PC).
XNA opens a way, but it will be hard to master it for beginners. The ones who pay will have xbox360 as additional platform to play with (but no network). Without paying, if some effort is made to obtain a correct PC graphic card the platform xp+xna may reveal itself interesting and a competitor to ps3 & xbox360.
The question is...
Is it possible to optimize GPU shaders in some unified way for PC graphic cards (with xp+xna) as far as it would be possible for xbox360 gpu?
Is xna acting as some abstraction layer (or direct3d) but in a way allowing really great optimizations?
I think there are things to study there and may be fishy stuff to dig out. That is, depending on the graphic card installed in a PC, things may be different when it comes to optimize a shader.
On the contrary if it's really possible to make all GPUs (for example ALL pixel shader 2.0 compliant graphic cards and the xbox360 GPU) work really hard in a very optimized way, then a really new friendly homebrew platform is born (mainly xp+xna because of club cost).
And maybe, the winner is not xbox360 over ps3, but xp+xna over xbox360 & ps3... (in that last case)
For xbmc users, however, they will have to stick to xbox1. At the moment it's the best and cheapest platform because it has been wide opened and codecs built in i386 world have been transported (not ported). With shaders, maybe these codecs may be ported but is it really a worthy effort (grab sources, licenses, port to shaders, etc... looks like a quite insane work, surely long to do...)?
However, with time, xbox1 will vanish since no longer manufactured. It's just a matter of time.
There is an opening there for a new i386 based machine that could fit near tv quietly and run xbmc.
(I will try to make xbox1 platform even better, soon, for openxdk developers... I try to hurry.
Maybe, I will have some ideas about how to build up a silent os-less pc in order to replace xbox1.
But the first step is to really master GPUs at lowest level. I'm working on it, hard. For those looking for xbmc and claiming they don't care about 3D, be aware that to reach some high video quality, GPU is needed, for antialiasing for example. It's, somehow, linked together. So we all need the true GPU power.)