I disagree that most headphones are crap, for the really good ones anyway. Frequency response is but one measurement of many. I have done a lot of distortion testing and my ears tell me that the distortion is lower in really good headphones than in most really good speakers. Especially intermodulation.
One great test that indicates much about many different aspects of performance is square wave testing. Phase response must be good, ringing must be low, to get anything that even resembles a square wave. Try it some time with great speakers. On all but the Quads you will see a total mess, not even remotely related to the square wave. The Quads give something resembling the best headphones which in the speaker world is considered amazing that they even display something remotely resembling a square wave.
The focus on frequency response is also a bit colored as the accepted methods of measuring for frequency response of speakers are based on experience, with the methods tailored to make them look understandable (i.e., some fudging going on). We could set up a protocol for headphones that would make them look good. For instance, even in-room response programs are made to ignore as much reflected behavior as possible. Otherwise, the data is a real mess. Drivers beam, so measurements usually try to ignore that and make big assumptions in excusing the different beaming patterns at the limits of the crossovers.
If I have to rely on one transducer to make an evaluation about a recording it will be headphones. Many mastering guys will tell you the same. If headphones were crap or bad this would not be the case.
Speakers can do things headphones cannot. But headphones can do things that no speaker I heard can.
Saying one is better than the other is not really a defendable position. Saying "I really prefer listening to (one or the other)" is valid and defendable.
