Auto Review token burn question

I decided to test the Auto Review setting in Codex for the first time in a while. I thought OpenAI may have improved it, but it might burn more tokens than ever. During a two-hour session of refactoring a Chrome Extension, auto review used 52% more tokens (API cost) than when manually approving actions. The actual work done with gpt-6.1-sol used $14.31 in API costs, while the Codex Auto Review used another $7.54. Is this normal? Is there anything that can be done to improve it?