Page level CRCs would reduce space / CPU overhead in our storage systems, but would introduce some additional complexity. It's worth exploring whether it's worth moving towards.
The main benefit of page-level CRCs over record level is it's trivial to discard them after verification after loading a page into the buffer pool. This would lead to both increased cache utilization since we don't store them in RAM, and better CPU utilization since we verify only on page-read.