I evaluate for a few different things in any given pair of headphones.
First, I look for resolution and clarity (the intangibles, in other words, that make a difference) by playing the Pain of Salvation live album 12:5. It's recorded to incredibly exacting standards and will really show off your gear's capabilities (and, by the same token, show problems in new gear). Very dynamic in every sense of the word, and thus a good starter test.
Then, I check how well the headphone handles layers of sound upon layers of sound by playing some Devin Townsend Band or Strapping Young Lad, usually DTB's Accelerated Evolution and SYL's City. If it can't keep up, it won't stay in my collection. This is a taxing test for the speed and bass resolution of the headphone.
Finally, I just listen to a bunch of songs I love for their various qualities:
Dream Theater's "6:00" for great drum placement aiding the detection of soundstage differences (same with Pink Floyd's "Time");
Lynch Mob's "Wicked Sensation" for its incredibly tight, crunchy guitar riffs to see how well the headphone handles tight, overdriven guitar;
Porcupine Tree's "Halo" because it's primarily drum and bass led but has great, countermelodic piano playing and heavily expanded chorus dynamics - you can examine the headphone's capabilities in the low and the high end in addition to seeing how well it can render the shift from a comparatively compressed rhythmline into a spacious, saturated chorus;
Opeth's "Moonlight Vertigo" because it's a beautifully recorded song and can tell you right away whether or not the headphones are any good for melodic death metal;
and a host of others. I put them through the paces, basically.