Its not just fabric it’s also liquids even when the PPU did both fabric and liquid and CPU skipped both the PPU was still faster.
There reviews where terrible as well the one from tomshardware was particularly bad with them blaming the problems with Havok physics on Ageia and they also used very old drivers. Most of those review used old drivers with known performance problems in Ghost Recon that have since been fixed. Why can we not get reviews with new drivers and games other then Ghost Recon?
How can we trust a review by someone who doesn’t even know the difference between Havok and Ageia and who use release drivers not any of the drivers that came out since. They used drivers about 5 versions out of date.
I won't trust Ageia's drivers simply because they said they fixed the Ghost Recon problems, the driver was tested (by Anandtech and others), and the performance was almost identical. Since Ghost Recon was just particles, it didn't exactly provide a good impression.
You may trust the review, you may not. The point of the software article was merely to point out that all the amazing physics you did see (the boxes, and vehicles) could be performed by a single CPU just fine. This isn't dualcore, quadcore, just a single core. The cloth is barely here nor there, and certainly not worth the price of a very powerful graphics card.
If anyone’s interested there’s some new vids up at http://www.ageia.com/physx/videos.html of BOS: Blood of Sahara where the fluid weapons have a big impact on the game (not to be confused with BOS: Bet on Solider which also use’s the PPU). There is also a vid of City of Villains.
Also look up Joint Task Force.
I looked at the videos, very impressive

However the Monster Madness video looked no better than the Xbox 360 trailer of the game, which as you know has no PPU. Most of the effects in the videos were basically particle systems with higher counts, and greater interactivity. Nothing that can't be done with a faster CPU, especially now that dualcore is coming in. I'm sure PVR-based GPUs would quite like these particle systems, I'm sure you could get a good 50x overdraw or more with some scenes.
However JTF is something else (video here:
http://video.google.com/videoplay?docid=6433791285344086653). Very impressive, and much more than just particle systems. However, I still have the nagging feeling that a dualcore CPU wouldn't have much trouble getting close to that sort of effect. Sure the extreme detail will be gone, but it would still be roughly right. Afterall HL2's physics aren't exactly 'bad', and they run on low-end CPUs.
The thing is according to Havok there hardware physics do not do gameplay physics. Do you really want none gameplay physics on the GPU over gameplay phsyics on the PPU?
Well, almost all games utilizing the PPU use it almost purely for pretty effects - so its not that different from the Ageia solution

Unless lots of people have an Ageia PhysX, its not an easy decision to have the PPU affect gameplay. You don't want to stop people playing your game, since you need cash to fund it. The big, big advantage of having the GPU handle 'fluff' physics is that the graphics card can offload the graphics and the physics from the CPU. Its like an extension of T&L/Shaders. The CPU can then focus on the core game itself, its quite elegant imo. Physics effects, graphics effects, all handled by a single card. Scaleable too presumably, so you get more and more particles automatically generated if you have a more powerful GPU (with attached physics in the GPU itself).
You don’t lose FPS you gain FPS. Please stop looking at reviews with the old drivers. I think the performance bugs where fixed in the 2.4.4 drivers or was it 2.5. Anything below 2.4.4 and older will be slow.
I have never seen a benchmark where the Ageia increases FPS. Since the PhysX creates more visual detail indirectly, the FPS will have to drop - the GPU is doing more work.