Has this solution been tested at scale? How many records can it handle before max CPU time exceeded? I'd like to use it for a cleanup process to avoid data skew, so I have thousands of records to move.