Performance drops after a couple of minutes of compiling

Thermal throttling. Check the sustained clock with a monitor running, then decide between undervolting, a lower power limit that trades peak for consistency, or a cooling pad. The consistent lower clock usually finishes builds faster than the sawtooth.

For a filesystem that checksums data, ECC protects the one place checksums cannot: memory between verification and write. It is not mandatory, and the honest framing is insurance whose premium is a slightly more expensive board.

Look at TBW rather than sequential speed. A consumer drive rated at 300 TBW dies years earlier than one rated at 1200 under the same load, and the fast numbers on the box describe bursts to cache, not sustained writes.

At ten times the size the bottleneck moves from the operation itself to what it competes with: memory, locks and the people who have to run it. The approach stays, the batching gets smaller and the schedule matters more.