KMeans provides less than K clusters

[komzpa]

Clustering 100 distinct points into 100 clusters gets 96 clusters:

select count(distinct cid) from 
(WITH
points AS (
    SELECT ST_MakePoint(x,y) geom from generate_series(1,5) x, generate_series(1,5) y
)
SELECT ST_ClusterKMeans(geom, 25) over () AS cid, geom
FROM points) z;

The larger K is, the bigger losses are.