Wingz-
"The GTS gets almost twice as many "marks" as a V5. Does that indicate that in real apps, the GTS will be twice as fast? (The answer: No)"
Well, Giants is in the 60% faster range with a GF2(or GF1 in many cases) and also running with added features enabled not possible on the V5(Dot3 for instance). In Sacrifice the GF2 is pushing higher poly counts and holding down FPS in the 30%-50% faster range most of the time. MDK2 when using the highest graphics settings in a non bandwith limited resolution the GF2 is around 250% faster then the V5(yes, that is supposed to say 250% and not 25%) though I haven't seen any numbers from the latest drivers which are supposed to be much improved there). Then we could get into a real app, something like 3DSM, Lightwave or the like where the GF2 is much closer to ten to fifteen times as fast as the V5
If you want to use 3DMark2K to measure the performance of games from '99 or earlier it doesn't work very well(Quake3, UT, Expendable, mainly the old benches that sites run to try and slant things in 3dfx's favor

). It seems that 3DMark2K's numbers hold up very well for year 2000 games.
jack20_00-
"It's not because the T&L unit, because Radeon and GTS's T&L perform level is extremely similar."
No, they aren't even close. I was hoping that the Radeon would have a strong T&L unit, but it very clearly doesn't. If you want to see the T&L power then check out scores at 640x480 16bit on anything pushing a decent amount of polys. Using some of the long standing industry standard benches such as Indy3D or ViewPerf the Radeon gets trounced at 1280x960 32bit many times by the GeForce1 SDR(they are not fill intensive benches, T&L mainly). We can hope that the RadeonII will fix the rather serious performance rift between nV and ATi, but one does exist right now.