можете да използвате функции на Windows като dense_rank :
select *,dense_rank() over (order by b) group_number from t order by b
така че можете да направите:
select * from (
select *,dense_rank() over (order by b) group_number from t order by b
) a where group_number<=2