Take advantage of your dual processor. OS X was designed to run on a dual processor and you're wasting away your G5. I would set it to 85%. Secondly, there's a simple conversion to figure out your latency. H/W buffer size divided by sample rate. If your session is at 44.1, 44.1 samples are in 1 millisecond. If your buffer is set to 512 samples and your session is at 44.1, that equals 11.61 milliseconds of delay. Now, for monitoring, LE is "double buffered". Your signal is going from your mic, i/o, processed, then sent to your monitors. So multiply your delay by 2. When you're recording, set your H/W buffer size as low as you can. When mixing, set your buffer as high as you can, so you can use as many RTAS plug-ins as you can. As far as your delay for your plug-ins go, only plug-ins like Maxim and Auto-Tune do that enough for you to worry about. That's because they seek ahead and make a graphical display for you in "real time". In TDM systems, the TDM plugins are processed last, after it's sent to the mixer and that's where you get that delay. In LE systems, its processed before it's even sent to the mixer, so it's no big deal. You do get a slight routing delay through sends and busses. The only way in LE systems that you can compensate for the delay is to zoom way in on 2 tracks and scoot one back by however many samples the track says it's being delayed.