I've been playing around with Aggregate Audio Devices since upgrading to 10.4. Awesome concept. Usually it seems to work well but sometimes not. I particularly like the idea of using Cycling 74's soundflower virtual device as part of an aggregate audio device and using it to route audio between applications in realtime (ie logic -> A MaxMSP creation -> logic). Or getting an extra lo-fi 2 channels of IO using my builtin interface with my firebox.
I'd like to know: What are your experiences with Aggregate Audio Devices? Are they reliable? What works, what does not? Do you notice more clicks and pops?
Here are my experiences on my powerbook G4 1.25GHz with 1.5GB RAM: First audio device listed is always the "main" one in the setup.
I tried Firebox + Built-in + Soundflower-16ch:
really cool possibilities, but it sucked up a LOT of CPU (presumably for running 22x24 channels and jitter compensating 3 interfaces). Occasional clicks and pops, even when running only one sin oscillator in MaxMSP... Promising for faster systems, but NOT so good on my PB.
Firebox + Built-in
I tried recording a performance in 6 channels with this config (4 from firebox's analog ins and 2 from built-in). It seemed to work ok when i tested it, but after leaving it on for 1 hour recording with just the 1st 2 channels from the firebox (recording friends who performed before us) in MaxMSP, I went to record out performance and it the built-in inputs did not work (nothing came through them. I restarted MaxMSP and reset stuff in AudioMIDI setup, but still i didnt get anything in the builti-ns except a nasty DC offset. Everything was setup right. I gave up and just recorded the 4 working channels since we were going on, but that sorta sucked.
Unfortunately (of course) aggregate audio does not work with Mbox
Figures. 
best,
Arvid