sondriver - The capability you are looking for is called Input Monitoring, and Sonar does have this ability. However, there is an issue called latency that effects the usability of this feature dramatically.
Simply put, latency is the amount of time that elapses between when you sing (or play) something, and when you hear the "processed" signal. Because the computer must take the incoming signal, apply the chosen effect on the fly and feed it back to the sound card, there is always somewhat of a delay when using input monitoring. If that delay is under 5ms, you generally won't notice it. However, if it is higher than 5ms, you will not be able to sing along with a pre-recorded track since your vocal will always seem to be *behind* the music (i.e., the delay that Ethan was referring to).
So what affects latency? First, you need a good sound card and (working) WDM drivers for it (read that, no SB Lives or Audigy's please). Next you need a pretty powerful CPU. Lastly, the amount of total tracks and the number of effects will also come into play. In other words, if you already have 15-20 tracks each with some FX, and you are trying to play those and record your vocal using input monitoring - well, fuhgadaboudit.
In my personal (and occasionally humble) opinion, Input Monitoring is an idea that is not yet ready for Prime Time. By that, I mean *most* people can not use it on their systems in the manner they want to.
I know I'll get lots of arguments that "it's working fine for me." However, in most of those cases, the users have state of the art CPU's, good sound cards, and even then, they turn off most of their prerecorded tracks to get it to work.
I can accept that it may be working OK for some, but for the general home recording hobbyist the hardware hasn't quite caught up with the software yet.