I'd assume that verifying that non-performing Googlers are actually working, and correcting performance issues, is much harder when they're remote. Most knowledge workers pretty much have to be trusted to perform their jobs, but that trust is much harder when you can observe neither process, progress, or results.

In most knowledge work - not just Google - you're fine as long as you're either putting in effort or getting results. After all, software development has plenty of risks, it's possible to fail to deliver for many reasons other than being lazy. But when you are neither putting in effort nor getting results, that's when you're at risk of getting fired, because the company starts to wonder why they're paying you. Remote work makes it much harder to observe the "putting in effort" piece.