AE CS4 on OSX 10.5 Very Slow, Makes System Unresponsive

Question

This is very frustrating, our company has been fighting this problem since we upgraded.  I work for a major broadcaster which uses AE for all our motion graphics work.  Since upgrading to CS4 we've experienced a team wide decrease in productivity.  AE CS4 renders slower than CS3 and in some cases will completely lock up the computer when rendering, making it impossible to do any sort of simple multitasking.  Even switching to the Finder and moving around windows is very slow.  Let alone try to use Safari or type into a TextEdit document.  We've tried everything, but details first.
We're running on last-gen 8-core 3.2GHz Mac Pro's.  Each has 10GB of ram and 1TB of HD space (mostly empty).  NVidia GeForce 8800 GT video card.  Runing AE 9.0.2.42)
We realize that 10GB is not enough ram to use all 8 processors at full speed.  So AE is set to use multiprocessing with a max of 3 cores with 2GB minimum with 2 GB in reserve for other applications.  All the work we do is with HD footage.  All footage is delivered in Apple Prores.  Any comp of medium or greater complexity will completely slow down the system.  When comparing CS3 rendering time to CS4, CS4 seems about 30-40% slower. When rendering CS4 will also completely lock up the system, switching windows is a complete pain.  We've tried turing on and off OpenGL in AE, but no change.  We're tried every combination of settings we can think of.   Rendering without multiprocessing is even slower and still locks up the system. Also in Activity Monitor is will often show only limited CPU useage (see attached screen).  You can see I'm running multiple aeselflink instances, but each one is no where near 100 and the system is still unresponsive!
Outside of rendering CS4 also is much more unresponsive when just working (not rendering).  Moving clips around in the composition window and navigating the timeline are sometimes a chore, with lots of 1-2 second delays before taking the action.  Or slow screen updates.
Any ideas what is happening here?  If is something with the change from 32-bit to 64-bit.  Will more ram fix it or is there something else going on?  This is becoming a very frustrating situations where we have to completely restart the computer and make sure no other apps are running before even trying to render our anything.  CS3 did not seem to have these problems, even with multiprocessing turned on at it's default settings.  We like many of the new features in CS4 and are practically required to use it because we get a lot of projects from vendors, but it's proving very slow to work in. 

Answer 1

I don't know how complex your comps are, but our are usually pretty complex.  Simple comps with just text and 1-2 layers of video usually go pretty quick, but as soon as you ramp up complexity it slows way down.  I somehow don't think that our graphics card is the problem.  Most of the plugins we use do not see any benefit from OpenGL and we don't use that many 3D layers.  In general we haven't seen a big boost in AE performance with OpenGL on or off.  All our footage sits on a super fast Terrablock system with at least 300-400mb/s sequential read speeds.  Then again it could be that AE just runs poorly on Macs.
Adolfo - we have done some quite extensive benchmarks with all our machines here.  Configuring both 10GB and 16GB versions (we have not tried higher memory yet).  We tested the same comp in both CS3 and CS$ with various multiprocessor settings.  Our test file's complexity was indicative of the kind of projects we do.  Around 30 layers, a mix of TGA sequences with alpha, QuickTimes with alpha (animation) and ProRes footage.  Many layers with ReelSmart Motion Blur.  A Knoll Light Factory lens flare.  A mix of layers with various blurs and color correction and some transfer modes.
As you can see CS3 was much, much quicker at rendering the project.  In some cases almost twice as fast as CS4.  Also, strangely CS3 ran better limited to 4 processors in both the 10GB and 16GB configurations. 
Benchmark Test File
CS3 Benchmarks
[note - CS3 has no options in the app for adjusting multi-processor settings.  All edits had to be made in the pref file.]
10GB System -----------------
AE CS3
10GB
4 Processors
12m 43s
[note - This was the most optimized setting for CS3 which gave the best performance for us.]
AE CS3
10BG
8 Processors
24m 8s
[note - Default setting in AE CS3 for multiprocessor.  System is completely unusable during render.]
16GB System -----------------
AE CS3
16GB
4 Processors
11m 54s
[note - This was the optimized setting for CS3 which gave the best performance.  Each instance of the program was using 150-180 CPU in the activity monitor so it must be using more that 4 processors with this setting.  Each processor used more memory than in the 10GB config.]
AE CS3
16GB
7 Processors
16m 15s
AE CS3
16BG
8 Processors
20m 13s
[note - Default setting in AE CS3 for multiprocessor.  System is completely unusable during render.]
CS4 Benchmarks
In general CS4 is about half the speed of an optimized CS3.  Going from 10GB to 16GB of memory gained us about 33% performance increase in the most optimized setting (4-6 processors at 2GB+ per processor).
10GB System -----------------
AE CS4
10GB total RAM
8GB reserved for AE (2GB reserved for OS)
3 Processors (2GB per processor)
24m 7s
[note - Computer response pretty slow]
AE CS4
10GB total RAM
8GB reserved for AE (2GB reserved for OS)
7 Processors (.9GB per processor)
28m 10s
[note - Computer completely unusable while rendering]
16GB System -----------------
AE CS4
16GB total RAM
8GB reserved for AE (8GB reserved for OS)
3 Processors (2GB per processor)
22m 57s
AE CS4
16GB total RAM
13GB reserved for AE (3GB reserved for OS)
4 Processors (2.33+ GB per processor)
18m 7s
AE CS4
16GB total RAM
13GB reserved for AE (3GB reserved for OS)
6 Processors (2.09 GB/Processor)
19m 1s
AE CS4
16GB total RAM
14GB reserved for AE (2GB reserved for OS)
7 Processors (1.85GB/Processor)
23m 4s
[note - This config made system very slow, left very little memory for the system.]
AE CS4
16GB total RAM
13GB reserved for AE (3GB reserved for OS)
7 Processors (1.59GB/Processor)
25m 50s