Initial testing shows that we might be a bit too strict when deciding to move to blocking writes. We'll probably want to introduce additional time for the timeout value to account for variance in workload and system performance.

Exact values are not decided yet, but this is a placeholder for that discussion.