It'd be hard to prove it, but in practice it gets noticeably slower at peak usage times after exceeding the cap and stays fast at off-peak times, which matches the expected result of deprioritization.