Behind the Tip of Efficiency: Uncovering the Submerged Threats of Jailbreak Attacks in Small Language Models