Most likely, unless Apple is using a new processor, this is the same as the Xserve. You can't just take a processor designed for 100 mhz bus and drop it in with DDR and say "Here you go, take it all." It doesn't work like that.
What Apple did is they took the FSB and brought it up to 166.67 mhz (if you want to be technical about it, there is no 167 mhz bus- only if you round it to be!). This is no different than how some overclockers on the other side make their CPU faster- by keeping the multiplier the same but increasing the bus. So, your comparison here:
Old G4: 133 bus x 7.5 multiplier = 1 ghz
New G4: 166 bus x 6 multiplier = 1 ghz
By increasing the FSB, you can get a speed increase because it is the FSB that controls the CPU, not the memory speed. Want proof? Take a look at the P3 and RDRAM and DDR. The P3 was running on a 133 mhz bus, and having more available memory bandwidth than it could use was a waste.
So why go to a DDR bus?
It's simple- it does benefit devices like high performance pci cards that do direct memory transfers. I dunno if this will make it clearer, but here:
_______________PCI Bus___------AGP Bus
CPU 1--\___________|______/
______(166)---System Chipset ----(266/333)---- Memory
CPU 2--/___________|______\
________________IDE____-----Firewire/USB/Sound
(_ = space, board didn't like whitespace in there)
So, the math works out like this:
Memory Bandwidth: Assuming a 166DDR bus (since the specs state either 133 or 166, I'm assuming that it runs at the same frequency as the FSB)
All numbers in MB/sec
Memory: 2660
CPU: -1330 MB/sec
64 bit PCI: -266 MB/sec
4x AGP: -2112 MB/sec
USB -1.2 MB/sec
Firewire -40 MB/sec
Sound -133 MB/sec
Gigabit -125 MB/sec
-----------------------------------
-1347 meg/sec
Now, you have to realize that those are max values, and you usually won't ever saturate the AGP bus, and only come even so close when playing games where you have to swap out AGP card memory with real memory, and Sound will never use 133 megs/sec, but you get the idea. In a SDR memory subsystem you would be cutting into the CPU's bandwidth immediately, whereas with a DDR system you can have many things going on without stealing bandwidth from the CPU.
So why did the Xserve DDR systems do no better than a MP Powermac?
One problem that is shared with both Intel and Apple (but not AMD) is that the FSB is shared by BOTH CPUs. That means that if both chips are doing 100% work, they have half the bandwidth the the chipset. So, any increase in the available bandwidth (which we'll get to next) is effectively halved, giving only around a 50 megs/sec advantage of a SDR system.
Which brings up the second point: Very few things are done independently of the CPU. So, even though a gigabit network card can have memory bandwidth independent of the CPU, it still needs to communicate with the CPU, using the CPU bandwidth and CPU time.
So, the main difference between this and the Xserve will be the FSB. And that WILL make a difference in benchmarks. They didn't need DDR, but since that's where the general trend is going, it's not a bad idea to implement it into current systems so that you can use the ram again in your new system. But, I repeat, without a major architecture revision in the CPU, it will NOT benefit from DDR as much as Apple would want people to think. The 166.67 mhz bus, however, will. _______________