Is there a performance difference between; • CSV Import using multithreading with 5 threads • Map/Reduce script using a concurrency limit of 5 Scenario would be updating a field on 20,000 custom records. That update would trigger a beforeSubmit function on those records that would update another ~5 fields on each record.

For custom records, there is a third option to mass update via a parent record, as explained in the SAFE Guide

With a CSV import, you have the option of disabling server-side scripts and workflows which could improve performance considerably depending on how many scripts/workflows you have on the target record. All scripts/workflows on the target record will run in a M/R context unless they've individually been excluded from that context.