Some things are legitimately very difficult to meter in a real-time way with adequate performance. Imagine if the S3 server had to do a DB query to lookup your account balance & limit on every HTTP request. That'll completely kill performance and availability (what if the billing DB is down? Should that DB now be replicated across all regions? etc).

The reason a lot of cloud alerts and metrics lag is because billing is done asynchronously by parsing logs way after the actual usage occurred. Of course, the real solution here is to just not bill for things you can't easily measure & limit especially when those things are super expensive, but cloud providers' C-suites have to eat & pay their bills somehow.