... to avoid timeouts, we should pull from each cursor in turn.

Special case - if the sort order is over the shard key - but in this case we'd have to detect chunk bounds anyway so probably only a small optimization if any.