The same way backups are most rigorously performed by people who've lost data, ECC is a non-negotiable requirement for people who've suffered slow data corruption via silent memory failures.
It surprises me that people are happy with 64GB+ builds of non-ECC, especially for NAS (ie. very long term storage, where corruptions probably wouldn't be noticed for years).
Periodically I look at replacing my small fleet of HP micro Gen8's, which use Xeons, have 4 x 3.5" bays (with proper h/w RAID1), but max out at a frustratingly low 16GB. They're quite robust, but because of their age - and HP - a horse veterinarian approach to component failures is usually indicated.
A Ryzen + ECC whitebox build is massively appealing, but almost everyone's build-out includes caveats like 'check the datasheet of the mobo and CPU' (because series aren't consistent), and about half the time a terrifying disclaimer to the effect of 'ECC is present / enabled in the kernel, but I can't tell if it's actually functioning properly'.