Why is MySQL using 4 bytes for each character when selecting something from a derived table when it has a concat() using a utf8mb4 charset?

Starting from this table with a utf8mb4_unicode_ci collation:

DROP TABLE IF EXISTS preferences;
CREATE TABLE IF NOT EXISTS preferences
(
    id      int auto_increment primary key,
    user_id int          not null,
    title   varchar(150) not null
) collate = utf8mb4_unicode_ci;

using these values:

INSERT INTO preferences (user_id, title) VALUES (1, 'Preference 1'), (1, 'Preference 2'), (1, 'Preference 3'), (1, 'Preference 4'), (1, 'Preference 5');

This query

SELECT CONCAT('"', GROUP_CONCAT(p.title SEPARATOR '","'), '"') AS title
FROM preferences AS p
GROUP BY p.user_id

with a group_concat_max_len=64 results in

But when using that query in a derived table, like so:

SELECT preferences_grouped.title
FROM (
    SELECT CONCAT('"', GROUP_CONCAT(p.title SEPARATOR '","'), '"') AS title
    FROM preferences AS p
    GROUP BY p.user_id
) as preferences_grouped

you'll get only 16 chars (64/4)

This is not happening when the concat is removed or moved to the primary query:

SELECT CONCAT('"', preferences_grouped.title, '"')
FROM (
    SELECT GROUP_CONCAT(p.title SEPARATOR '","') AS title
    FROM preferences AS p
    GROUP BY p.user_id
) as preferences_grouped;

So it seems that adding the concat in combination with a derived table ànd a utf8mb4 charset causes the byte count for the chars in the result set to be quadrupled, meaning my group_concat_max_len should really be quadrupled too.

1. Why is that?
2. Can I do something using COLLATE to fix this / change this behavior?

Note: When the preferences table has a non-utf8mb4 charset, like the default latin1_swedish_ci collation, then there's also no byte-quadrupling going on.

Check the fiddle here: https://www.db-fiddle.com/f/hQD6r3pE5pmgR3n2XA7VtX/5