Shouldn't it be 250 - 100 ms (rather than plus)? The person who was talking is silent for that extra 100ms while they wait for their words to reach their peer.
What I imagine is that when I'm listening I need to wait the regular wait time plus network lag because the speaker may have resume speaking. The speaker needs to resume speaking before 250 + 100 has passed so that others don't start talking before the speaker resumes.