The "lfs mirror resync" command currently computes parity single-threaded. For systems with many cores, parallelizing the computation (e.g., computing 32 different RAID sets in parallel on 32 cores) would significantly improve resync performance.

Note: The I/O side is already well-parallelized internally by Lustre for direct IO, so the bottleneck is computation, not I/O. This optimization becomes less critical once immediate writes/updates land (reducing the need for resync), but is still valuable.