The Radeon definately helps unlock the potential of the Athlon 64. Sites that test with a GeForce FX 5900 or 5950 Ultra show very little difference in gaming apps between the Athlon 64s and P4s, leading me to believe the video card is a bottleneck.
Tom's Hardware also seems to be doing something funny with aggressive memory timings on the P4 systems and slow memory timings on the Athlon 64 systems. Look at their test setup for the 3400+ review:
P4s -
Asus P4C800-E Deluxe Rev.: 1.02
Bios: 1011 BETA 006
4 x 256 MB CL 2.0-2-2-5 (200 MHz)
4 x 256 MB CL 2.0-2-2-5 (133 MHz)
Athlon XPs-
NVIDIA nForce2 Ultra
(Socket 462) Asus A7N8X Rev.: 2.00
Bios: 1006
2 x 512 MB CL 2.0-3-3-6 (200 MHz)
2 x 512 MB CL 2.0-2-2-5 (166 MHz)
2 x 512 MB CL 2.0-2-2-5 (133 MHz)
Athlon 64s-
VIA K8T800
(Socket 754) MSI 8KT Neo (MS-6702)
Bios: 1.0 Rev.: 1.0
2 x 512 MB CL2.0-4-4-8 (200 MHz)
The Athlon 64 is stuck with the relatively mediocre MSI K8T Neo, and the memory timings are 2-4-4-8 versus 2-2-2-5 for the P4s. This certainly doesn't help the Athlon 64. (It's also possible that the MSI board gains performance with newer bios releases than 1.0, which is the first final release, even though 1.1 is available.) The P4 has the very nice (but expensive) 875 based Asus P4C800-E Deluxe, which is an extrememly nice $200+ dollar board. The K8T Neo is a $120 board.
There's also one more thing which will really hurt the Athlon system - Tom's used a PCI slot Serial ATA controller card on the Athlon 64 in Raid 0 mode, rather than the very fast VIA integrated Serial ATA controller. This means the Serial ATA controller is bottlenecked to 133 MB/sec by the PCI bus itself, rather than the 533 MB/sec offered by the Vlink for the Serial ATA ports in the VIA southbridge itself. In a Raid 0 configuration, this is a serious bottleneck, as the 133 MB/sec is shared by ALL PCI devices, not only just the Raid controller. A good Raid 0 array can easily transer 100 MB/sec +. The Intel board used the Intel Serial ATA ports onboard in Raid 0 mode, meaning the Raid 0 array was not bottlenecked by the PCI bus:
Hard Drive (AMD System) FastTrak S150 TX2plus (Bios: 1.00.0.30)
2 x SATA Maxtor 6Y080M0 (Raid 0)
80 GB / 8 MB Cache / 7200 rpm
Hard Drive (Intel System) Intel FW82801ER ICH5R
2 x SATA Maxtor 6Y080M0 (Raid 0)
80 GB / 8 MB Cache / 7200 rpm