[Edit: Here is how I think it works, from my brief 2-day usage of it.
Comet: persistent connection between client and server.
First client connects to the server and creats a session. Server creates a "channel" with a file-handle/chan-id. Subsequent clients follow a URL to join the channel.
Without the live editing, you can see how any write in one client's textarea can be sent to the server and broadcast to all the channel members.
To allow for live editing, server keeps a "buffer" for the channel, initially empty.
At the start, the server sends a test packet to each client and sees how long it takes, for timing purposes, call the time it takes to get a PONG response L for lag. So in the future, when a client sends a text snippet, it's timestamped, and the server subtracts the lag to get the actual sending time.
Along with timestamps, each text segment is also ID'ed by user, this allows the rendering engine to assign colors to the text.
Just wild speculation after giving this no more than 5 minutes of thought and another 10 to type it.