On three puzzles in robust control