I've been having a hard time foreseeing the power of ATI's upcoming R600 GPU.
The R600:
64 dual-threaded unified shaders
GDDR4 at about 2GHz DDR
512-bit memory controller

The G80:
128 unified shaders
GDDR3 at 1.8GHz DDR
384-bit memory controller

Okay so we all know (you shuld know if you intend to reply with a good post) the advantages of a unified shader language. Having lots of shaders helps
The G80 has 128 stream processors whereas the R600 has 64 of them. However, the ATI shader will be able to process two threads at once, which theoretically makes it able to process 128 shader operations at once, like the G80.

Now let's go to the RAM.
NVIDIA has done it right by using cheaper GDDR3 but upping the interface to 384 bit. It enables the RAM throughput data performance to a massive 86.4GB/s. That's FASTER than the X1950XTX's despite that the latter uses GDDR4. Now the R600 uses a 512-bit interface AND GDDR4.

Now taking into account that the future games are VERY shader intensive (Oblivion runs like butter on the 8800GTX's 128 stream processors and stutters on the X1950XTX's 48/8 shaders), which video card will be faster in the upcoming titles?

It's easy to say that the R600 will be faster since it can do 128 shader operations like the G80 AND it will most likely have a higher memory bandwidth, but remember that even the 7900GTX has ONLY 24 pixel and vertex shaders AND is able to compete with ATI's X1900XTX (48/ in many games.