Reggie, when you're talking over the phone, the phone is just sending a real-time analog signal over the lines. There actually is a delay between you speaking and the receiver hearing, but it's so tiny that you'd never really perceive it--electronic signals and light travel incredibly fast, but they are not instantaneous. I've never actually tried it, but I imagine you could sing a song together with someone on the other end and neither of you'd ever notice the difference.
When you're doing PalTalk over the Internet via a regular land-line dialup, what you say is converted into a digital signal by your computer; then it's converted back into analog by your modem (not the actualy sound of your voice, but an analog encoding of the digital stream, which has already now been divided up into small, manageable chunks called "packets"). This step doesn't happen, by the way, on my cable modem connection, but there is a conversion process of some kind at the cable modem anyway--I'm not up on the tech there.
Your ISP's modem converts the analog version of those data packets back into digital. From there the packets bounce around the Internet via routers, until they gets to PalTalk's servers, and then the whole shebang happens again in reverse as your voice comes chugging our way over the wires via varying formats and technologies.
Also involved is distance; your mention of satellite phones is apropos. Like a land line, a cell phone connection doesn't cover enough distance for you to consciously notice the delay. No avoiding it with satellite trasmission of any kind. Some of the musicians at the London and Philadelphia(?) ends of Live Aid were very disappointed to find that they wouldn't be able to sing with each other across the ocean. (Arthur C. Clarke in one of his novels depicts the befuddlement of political officials who can't understand why real-time communication between Mercury and Jupiter isn't possible.) In data networks, though, "distance" doesn't mean the physical length of the lines connecting you, but the number of "hops" between various pieces of connected networking equipment.
So this situation is much more like a satellite phone than a conventional one, although latency is built in to the process for different reasons. The first element of latency is physical distance--tiny in itself unless you're going via satellite. The second is logical distance: the number of hops. The third is processing time on every computer, modem, and router along those hops. The fourth is bandwith. I'm not sure I completely agree with Steve--until they're ever able to come up with a way to get around the lightspeed limit in communications (don't laugh; some day we may find it's possible with tachyons), and build unbelievably fast routers with incredibly fast and reliable pipes between them, I don't see real-time sing-togethers being a possibility. In short, a quantum-leap in technology. I won't say never and be made a fool, though.
Now, more than one mic being open at a time is another matter entirely, and trivial--NetMeeting does it, and PalTalk does it in one-on-one chat. But true real-time simultanaety (sp?) is another matter entirely--using NetMeeting through an MPCS server on our own campus, over a 100 megabit ethernet network, there's still noticeable latency.
In short, apples and oranges.
Chris, networking guy