Let's clear the confussion. The Hypertransport tunnel speed has NOTHING to do with the kind of memory used. You are still thinking OLD CPU architecture, where the only data exchange path from the CPU to the world was the FSB. Now, for the AMD64 chips, you have 2 paths to exchange data:
One of those paths is the memory controller, integrated into the CPU. The other path is the HT link, intended mainly for peripherial access and video card access (AGP, and soon PCI express). Both numbers are independent from each other. The fastest memory that can be used depends on the CPU itself, based on how high the integrated memory controller can scale.
Both numbers seem to be related because some motherboard implementations link them through a multiplier (WRONG approach, but I assume it saves money.... ). Both numbers should be totally independent, and I hope NForce3 250 fixes it (yes, having the choice to change the memory speed without touching the HT link speed)
Hence, you can have combinations of a extremely fast HT (1 GHz each direction, 2 GHz total, 8 GB/sec) coupled with low memory bandwidth (mobile Athlon DTR socket 754, DDR333 single channel, 2.7 GB/sec) On the other hand, you could have a very "slow" HT tunnel (ftake the speed of the nVidia NForce 3 Go 120, used for the transmeta efficeon, carrying no more than 1.6 GB/sec... 200 Mhz maybe?) coupled with a CPU that has potent memory access (Athlon 64 FX55, socket 939, DDR500 dual channel, 8 GB/sec).......
The memory can go as high as the CPU memory controller allows, and the BIOS must be aware of such capability. The HT tunnels is used ONLY for peripherials and AGP access... that is why the K8T800 performs better than the NForce 3 150 in workstation applications (intensive on the AGP bus).
Alex