A Penalty Approach to Linear Programs with Many Two-Sided Constraints