Three seconds is an astounding amount of delay. After more than about half a second delay it really begins to impinge upon conversations. That said, the delay comes from several things, such as encoding and decoding, packet size, and buffering. None of these are relevant to whether or not the information is retained. In other words, the call could be recorded with no perceptible delay, or not recorded with very long delays. The act of recording does not introduce any delay whatsoever.
You missed my point:
Recording must occur to create delay.