More chunks doesn't mean that it's more accurate. There is a balance. Retrieving unnecessary chunks gives the LLM irrelevant information and can confuse it and can degrade the answer quality. Retrieving too many chunks can introduce noise and unnecessary information, which will make it harder for the model to identify the most pertinent information. It can also degrade the answer quality.