I agree a compressor/limiter and perhaps some other effects would be nice to have already in the recorded track, without the original signal being distorted. If Sonar could do that, great! You could also save some processing power by using an external, and perhaps better compressor?
So far so good. But remember compression diminishes your signal/noise ratio. So why not just record a 24 bit signal, adjusted so as not to distort anywhere in the signal chain? You can add compression later or postrecording for monitoring, if for example a drummer needs to hear compressed cymbals while playing for better performance. If you use a compressor to avoid distorsion you are effectively just reducing your 24 bit dynamic range to something a lot less. So don't fool yourself into thinking you have 24 bit quality, just because a prerecording compressor prevented overflow.
As to the second issue, latency and recording vocals, I think for those with a slow computer/soundcard you could get something useable with, say, 10ms latency. If your card can route a direct dry signal to the singer, do that. That will be no latency, but also no reverb. Then add 100% reverb with 10 ms latency to the track being recorded and mix that with the dry direct signal, until the singer is happy with the sound. Later you can reduce the reverb when mixing the laid down tracks. I realize this work-around is no good for guitar effects where you usually don't want anything left of the original signal. But for vocals, I believe it could be useful.