Currently were running CS4 on MacIntel connected to Xserve Leopard.
Using AE CS4 and Particular we noticed that rendering a project with Particular (3 layers)only takes 2min32sec on the network (saving QT to Xserve).
However, we took the 3 layers and converted it as two QT clips (one with alpha the other as a bg clip).
This version of the same comp (using rendered QT saved on Xserve) takes 6min20sec to finish.
This was all done on a Mac Pro 4,1 system running SL.
Weird part too is that on an older Mac Pro 1,1 (same app/os) using the above settings, the render times were opposite, 18 min for the particular layers, 8 min for the imported QT clips.
This makes more sense since the Particulars had to be calculated as oppose to the QT clips just read frame by frame.
I know this all sounds ticky-tacky but we have a ton of comps that are templated and would like to lower the render times as best possible :)
So to sum this up, your computationally-bound renders (Particular) are faster on newer Macs than older ones, as expected.
Your IO-bound renders (compositing QT movies) are roughly the same on the newer Macs and the older ones.
That indicates to me that your IO is the weak link. How are you connected to the XServe? Ethernet? Fibre channel? Does it speed up if you composite the footage from direct-attached storage on your workstations instead of from the Xserve?
Try the AJA System Test from your workstations on the networked drive. (You can use the "Read existing file" test on one of the existing QT pre-renders). How fast is the access?
Principal & Designer at Keen Live
Motion Graphics, Widescreen Events, Presentation Design, and Consulting RenderBreak Blog - What I'm thinking when my workstation's thinking
Creative Cow Forum Host: Live & Stage Events