Real time, as in a 1:1 ratio between the time it took him to write it and the time it takes for you to see it played back. Not as in live, where you see it as he's typing it.
I'm still waiting for XEP-0301 to become ubiquitous (which transmits key delays with text, and would be closer to what the GP thought he saw) - I really miss this feature in chat systems.