Guys, this discussion went quite offtopic, but maybe you missed one
fact that Cyril may no know about.

In postgres, setting N columns to NULL is just N bits of physical writes.

So the overhead of

INSERT INTO t1 ( id, data1, data2, data3, ..., data100 )
VALUES( 12345, 'the only non-null data', NULL,NULL, ..., NULL )

is not so terrible.

In regard to UPDATEs, we _have_to_ specify all columns anyway;
and DELETEs are handled by PK value.

So there's not much to optimize / without making a revolutionary changes /


-- 
Filip Rembiałkowski