That's because simple questions in Bard only generate like 200 tokens per answer. The latency is more noticeable for longer answers.