1) It seems to me that with 4X FSAA, the cards are on par. Each one wins in different benchmarks.
I don't think so. Looking at Tomshardware and Hardocp, the GeForce FX wins until the resolution is cranked up too high (where both are below 60 fps...). This, to me, looks like a clear driver issue. It doesn't appear that the cards are CPU-limited, and the higher resolution would be easier on color/z-compression. But, regardless of whether you think it wins or not, it is impressive how well it does with less bandwidth in 4x. The Codecreatures benchmark is also very interesting...the FX actually improves its lead with FSAA enabled. My first guess would be that this is a memory granularity issue. The Radeon 9700 Pro uses 4 128-bit memory controllers, while the GeForce FX has 4 64-bit controllers. While it has less total bandwidth, it should still be better when accessing many textures at once. I don't really know for sure what this means for future games, but I think it is worth noting.
2) Based on rumors, NV31/34 won't have any clock-speed advantage over the Radeon 9500/9500 Pro, that the FX has over the 9700 Pro. Furthermore, the 9500 pro will have an immense fill rate advantage, due to its 8 pipe design.
Well, we'll see, but I doubt the 9500 Pro will drop much in price, since it's basically a 9700 with a 128-bit memory interface. I would imagine that ATI will try to get it phased out as quickly as possible.
3) And again, qualitywise, 4X FSAA on the GeForce is not comparable to 4X FSAA on the Radeon.
Yes, that's true, unfortunately. This is something that people have been complaining about nVidia for a long time now...hopefully nVidia's FSAA will be fixed soon...but, at the same time, right now the FX is quite a bit better than the Radeon 9700 for older games with alpha tests. This doesn't make a difference to me at all, but I suppose some might think it a plus.
4) You mention Heirarchical Z being disabled on the 9500 non-pro. Yet no consideration about all the "disabled/missing features" on the NV31/NV34?
Yes, that's true, I suppose. It's just that in the past nVidia has been
more likely to include the memory bandwidth-saving features in the lower-end parts than the higher-end parts. That is, look at the GeForce4 MX. It has the FSAA/anisotropic of the GeForce4 Ti, but not the programmability features. We essentially know that the NV31/NV34 will have the core programmability of the FX, but it seems likely that the memory bandwidth savings stuff would be there as well. I suppose you're right, though, this will determine whether the NV31/NV34 will perform very well or just decent (for the price). It will be very interesting.