Does capitalizing letters, using "*" chars, or other similar strategies to add emphasis actually do anything to LLM prompts? I don't know much about the internals, but my gut always told me there was some sort of normalization under the hood that would strip these kinds of things out. Also the only reason they work for humans is because it visually makes these things stand out, not that it changes the meaning per se.

Yes, upper and lowercase characters are different tokens, and so mixing them differently will yield different results.