AFAIK, that's not really how Google's system works. Yes, managers will give numeric performance scores to people on their team. And yes, those raw scores will almost certainly be adjusted. But it's not as crude as simply normalizing the scores to fit a given curve. Instead there's a process to try to make sure that one manager's 3.5 is the same as a 3.5 of the other managers in the same area and the same office, by looking at what they actually accomplished. Then the scores would be calibrated across progressively bigger and bigger units, until everyone is more or less on the same scale.

If it turns out that one manager is giving consistently higher scores than another and can't support it, then it seems quite reasonable and fair for there to be an adjustment down.

This is just my understanding of the process, based on comments by people involved in various stages of it, many years ago. Maybe I misunderstood, or maybe things have changed. It wasn't a perfect system either, people with a bad manager who can't argue their case could still get screwed. But despite all that, the calibration scores never felt like an adversarial process. Certainly nothing like some of the MS stack ranking horror stories.