I've pondered this question before because I have two musician brothers who live in different parts of the country. From what little I know of the internet, I think true "real time" is probably not possible. Internet communication is a bit of bucket brigade process, so time delay is just a part of the landscape. It may be short enough for a telephone conversation but for music I don't think it could ever work. That said, anybody who worked with Audacity when it first came out knows that digital recording is never truly real time. There is always a time delay between the outgoing sample stream from the DAW to the D/A converter and the incoming sample stream from the A/D converter. One of the things your DAW does for you (that Audacity didn't!!!!) is keep track of this offset and time shift the incoming stream to keep it in synch with the previously recorded tracks (so you don't have to do that by hand like in early Audacity). Seems to me something like this same process could work over the internet. I'm not aware of any software today that can do this, but i could imagine an application on your end that sends out the cueing material along with a clock stream of some sort. On the other end another application would receive and play back the cue and then collect your singer's vocal, attaching the appropriate clock data to the samples on their way back to you. Your application would recieve the sample stream and record it in your DAW with an appropriate time shift based on the clock data. The only other difficulty might be bandwidth. 24 bits at 44k or 48k sample rate is probably beyond what can be squeezed through the internet "pipe". This could be overcome with buffering on both ends. You confirm your singer is ready and then hit record. Your application begins sending the cueing data. Based on the lengh of the song, the application on the other end would store up whatever fraction of the song necessary given the data rate before beginning the playback for cueing the singer (kinda like TIVO

). Same deal on your end: there would be a buffering delay before you would begin hearing the vocal coming through (synched based on the clock with the cueing audio). It would be as if your singer were just in the next room except that there would be a short pause between when you hit the record button and when you beginning hearing the audio and her singing.
I'm pretty sure this would be possible, but it would need a pretty good programmer to put it all together. Whether it could be integrated with an existing DAW, I am not sure. Existing DAW software already manages the short time delays between playback and record, but these delays are in msec. I suspect with the buffering needed for the process above, we're talking 10-15 seconds, maybe more. Could the DAW be fooled into thinking you just have a really, really, slow audio interface? Maybe. I don't know...
A more practical suggestion is to render your cueing as a .wav or high quality MP3 and transfer that to your partner by FTP. She'd record her vocal and transfer just the vocal .wav file back to you the same way. Since you're not moving the entire mix back and forth this really shouldn't be very cumbersome.
J