I wonder how can you conclude that the plan chosen is
“as if the table contained only one row ?”.
A cartesian product occurs in any case when you use two
tables in the FROM clause, without joining them in the
WHERE clause, no matter whether you have statistics collected or not.
This depends solely on the SELECT statement itself.
Optimizer plans may indeed change when statistics change,
but they are generally supposed to improve, but there may
however exist cases where there are still things that the
optimizer can’t know (for example, not having histograms
where they could be helpful).