Ok, lets clear some stuff.
First, YES, you dont need a physx card, but you dont need a 8800GTX, 9800 nor G295 either. We get them because they give us better picture quality, more resolution, more effects, and so on. So, if you want some extra eye-candy the answer is YES, GET A PHYSX CAPABLE CARD. Remember that every day more and more games are using Physics.
Second, NO, a 9500 does NOT have the processing power of an Ageia's PPU. As a matter of fact, only G200 series are more powerful than the Ageia's PPU for Physx. You might find some synthetic tests giving a 10-15% higher score improvement when comparing the PPU vs "an nvidia sub G200 series card" (ex: 9600GT) but, in real life gaming, they get far behind (with differences up to 20 frames over 50).
Third, REMEMBER one is dedicated/discrete, the other, NOT. Ageia's PPU its going to give you better performance, because its solely used for physx. Any gpu used for both will result on more stress for the card, as this one needs to process complex physx calculations + Rendering.
Fourth, There are patches. There are patches for everything. From game patches to use physx (not that many) to Patches for Nvidia Physx drivers. So there's a big chance you would be able to run physx on a system with ATI. God, there's even a onboard chipset solution coming to the market soon, where you can run a Crossfire + SLI system, lets call it CrossSLI system. Where the chipset balances the nvidia card with the ATI. And guess what? the best performance combination is not a pair of 5800's in crossfire nor a pair of G280's in SLI, ITS A 5870 + G280 CrossSLI!!!
Fifth. While ATI is working on ATI Stream, and worked before with Havok, and nVidia have Cuda and physx, the future isnt certain but ... many people believes that the future is OpenCL. Which runs on Cuda capable cards, and ATI Stream capable too. So basically most of the future options (Physx, Cuda, ATI Stream, Bullet, would be implemented through the open platform OpenCL).
Sixth and last, NO, Software Physx and Hardware Physx are not the same. Im going to answer this with an example, its easier. I own Mirror's Edge and an Ageia PPU. Ok, there's a scene, right after the main character talks to her sister for the first time, where you are chased by the police. Well, the map is a corridor with 3 or 4 half columns, that have an exposition glass each one. Then a wating room with a couple of glass walls you can go through and finally a long room with a big glass wall spliting the room in half, where you get shoot at through those glass panels. The panels break into pieces and fall. My system specs are AMD Phenom II BE 950 Quad Core @ 3.1ghz. 4GB Ram. ATI 4830 512. Ageia PPU PCI with all the Patches and drivers applied (Asus, nVidia, GenL, Game patches, ... etc). Runs Crysis DX9 all on high (autodetect). And the rest of the games even smother (Gears of War and COD: Modern Warfare 1). Im trying other Physx games as well, warmonger, cellfactor, UT3 Physx mod, and saving some money to get Dark void and ... another one I cant remember now). So I might update this post with results as soon as I get those games.
Ok, first run, without physx. Frame rate about 50 frames though the whole scene. When the shoot the glass panels, they break into small pieces and fall down to the floor. It is an animation so, mainly, it looks always the same. Glass pieces disappear during falling, and as soon as they touch the floor, nothing is left. Bullets make holes on the walls, and ... thats about it.
Second run, software? physx. Im not completely sure, but the hardware physx driver was not working. Initial frame rate 50. As soon as they shoot the half-colum's glasses frame rate drops to about 35-40. Everytime glass looks different, and it bounces up the floor, then falls again, some pieces bouncing again and some not. Most of the glass stays on the floor as you keep running around. If you kick it, it spreads over the surface. When you go across the glass panels on the waiting room, the debris drops the frame rate to about 20 frames. Finally when you get to the long room and the bullets break all the panels you get 2-4 frames. When they shoot the walls, bulletholes appear as well, but you get wall drops as well on the floor. Nice effects, not a huge difference, but NICE altogether.
Third run, hardware physx. Same as before, lots of effects. 45 frames per second from the beginning till the end. REALLY COOL
. Usually harware Physx gives you, more effects than NO Physx or Software Physx, better framerates than software physx, but a bit worse than NO Physx (example: No Physx 36 frames, Hardware Physx 30 frames, software Physx 5 frames).
Please correct me if Im wrong. Thanks.
PS: I dont post links as Im a heavy reader, and I dont collect the links of every page I read. But I could try to find some of them if necesary.