puoi usare funzioni di Windows come dense_rank :
select *,dense_rank() over (order by b) group_number from t order by b
così puoi fare :
select * from (
select *,dense_rank() over (order by b) group_number from t order by b
) a where group_number<=2