If you want some numbers, some data they released years ago are that about 80% of validation attempts fail, and closer to 85% over IPv6. Since that time they've made some efforts at blocking "zombie clients" earlier in the process that just keep on retrying forever even though they haven't worked for years. In general I think the problem is with systems that retry too often (since they don't have an administrator paying any attention) rather than systems that don't retry often enough.
Well, the bandwidth and computational costs from trying again are much more than the costs of just making and tracking a new token, so I think it makes a lot more sense for the server to only retry (and with a new token) if requested to, and not just in the hopes that it might work eventually.
And as we keep saying, having an actual automated system, that can more smartly tell when to retry, and when to alert an administrator that multiple attempts don't seem to be working, really makes more sense from a wholistic perspective.