Strange result with "SELECT ... ORDER BY random() LIMIT 1" and JOINs

Hello,

Consider the following tables and data:

----------------------------------------
CREATE TABLE color (
     color_id integer PRIMARY KEY,
     color_name text
);

INSERT INTO color (color_id, color_name)
VALUES
     (1, 'red'),
     (2, 'blue'),
     (3, 'green'),
     (4, 'yellow'),
     (5, 'grey'),
     (6, 'brown'),
     (7, 'black'),
     (8, 'white'),
     (9, 'white with wooden panels')
;

CREATE TABLE car (
     car_id integer PRIMARY KEY,
     car_name text
);

INSERT INTO car (car_id, car_name)
VALUES
     (1, 'Toyota Matrix'),
     (2, 'Mazda 3'),
     (3, 'Honda Fit'),
     (4, 'Ford F-150'),
     (5, 'Chevrolet Volt'),
     (6, 'Audi A4'),
     (7, 'Hyundai Elantra'),
     (8, 'Nissan Versa'),
     (9, 'Buick Estate Wagon')
;
----------------------------------------

This query yields unexpected results (tested under 9.2.4):

SELECT
     s.car_id,
     s.color_id AS subquery_color_id,
     co.color_id AS join_color_id,
     co.color_name
FROM
     (
         SELECT
             ca.car_id,
             (
                 SELECT color_id
                 FROM color
                 WHERE ca.car_id = ca.car_id  -- dependency added to
avoid getting the same value for every row in the output
                 ORDER BY random()
                 LIMIT 1
             ) AS color_id
         FROM
             car ca
     ) s
     LEFT JOIN color co ON co.color_id = s.color_id;

We can see the equality defined in the LEFT JOIN does not hold true for
the subquery_color_id and join_color_id column aliases in the output.
EXPLAIN also shows that the subplan for the inner subquery used to pick
a random row from the color table appears twice.

I don't really understand what is going on there, the result appears
incorrect to me. The following page seems to offer some explanations as
to what is happening:
http://sql-info.de/postgresql/postgres-gotchas.html#1_6 . Indeed, adding
OFFSET 0 to the "s" subquery solves the issue.

Can somebody shed some light on this topic? Is this behaviour correct or
should it be considered a bug?

Thanks!
Etienne