If they are literally sending the entire transcript and asking for a summary, wouldn't almost all of the tokens be cache hit?

Maybe. But the 16.4% is from Claude Code's own cost_usd estimate, so cheap per token still adds up when it's 1 in 5 requests.