Yeah, it's not about adding the column; I know that's instantaneous. I'm talking about the UPDATE you'll have to issue to populate the column once it's been added — unless you're okay with it staying NULL (or doing some sort of "lazy loading" is viable in your specific scenario). If you try to do that for the entire table in one go, you're contending with concurrent writes, incurring a lot of IO at once, and as much as doubling the disk size of your table.

This way, you're hitting a limited number of rows per iteration, significantly softening the IO impact (granted, at the expense of wall-clock time); the NOWAIT fails fast on rows that have write locks out against them when you try to grab them; and (if the column you're populating is indexed, obviating HOT updates) leaving a radically smaller number of dead tuples — particularly if you up-tune autovacuum on the table in question while doing this (though that can mitigate some of the reduction in disk IO).