The updates wouldn’t be in place anyway because of the Postgres mvcc approach, right?

They might be if the columns being updated weren’t indexed [0], but since it sounds like at least one is, no, not in-place.

Though interestingly, your comment on BRIN indexes is quite relevant, as that’s the one type of index that HOT can still work with.

[0]: https://www.postgresql.org/docs/current/storage-hot.html