You would be much better served looking at reviews of card X where the synthetic marks you are interested in are implemented. Many users that bench these synthetics do not use the same methods, or "optimize", others outright cheat.
3dmark05 is not the best gauge for performance either, look at these links from a review. It shows 3dmark05 at a res many play currently, 1280x1024, then that same res with 4xAA/8xAF. If you were to look at the 3dmark2005 score with 4xAA/8xAF alone you could assume a 9800XT runs just about the same as the 6800. Now go look at the second link and the green bars on the chart for 4xAA/8xAF 1280x1024 in Far Cry. While you are there compare the synthetic marks/gaming measurement with 4xAA/8xAF of the X800XT-PE vs the Ultra as well. Different conclusion isnt it?
For added fun go seach 9800XT and then 6800 AGP at pricewatch. Compare prices.
To clarify my earlier point. If a fellow buys a 6800 then installs the drivers and benches 3dmark05 he will be using the driver defaults which are quality and AA/AF at application preference. If he compares that against another user who has high performance set in his driver and AA/AF to off with all his unnessary windows background services off as well, the numbers will not match and he will wonder why his card 'sucks' compared to the other guys.
You can not use a bench as a baseline to establish performance for cards if users keep changing the variables. You can add to this that you also have users that implement outright cheats like nulling shaders.
Thats why I suggest reviews. You have a more uniform controlled set of variables.