Disagreements in Reasoning: How a Model's Thinking Process Dictates Persuasion in Multi-Agent Systems