I am writing this just based on my experience as I have seen thousands of 4500 CPU cases.

When you look at this graphs the 72hr average gives you a reliable baseline average which in your case is a consistent 20%.

The *** means spikes and of course your switch will hit 100% in a single cycle (remember there are millios of cycles per second)because it will try to do all of its tasks at once. It just takes one busy cycle to reflect a *** in the high 90's.

The Cat4k Mgmt HiPri and Cat4k Mgmt LoPri are normally the higher processes.

You can use "sh plat health" to detail the two processes mentioned above. Keep an eye on the first two columns, the first one is the target % and the second is the current %.

You want to have your actual % lower than the target.