Overclocking LLM Reasoning: Monitoring and Controlling Thinking Path Lengths in LLMs