I usually do a lot of testing and change it up based on what purpose I wish to use the headphones for. Some stuff I'll test pretty much everything with though, either for informational purposes or for kicks.
Gershwin's Rhapsody in Blue, jazz band arrangement with Michael Tilson Thomas and the Los Angeles Philharmonic. - Detail. I don't play the entire piece as it's over 15 minutes, the big thing I'm listening for is the clicking of the clarinet's keys at the very beginning and how the sense of space changes when the full brass section comes in.
Copland's Fanfare for the Common Man with Leonard Bernstein and the New York Philharmoic. Mid/Treble balance. With this I'm listening for the relative volumes of the horns and the trumpets.
Mozart's Great Mass in C Minor, K. 427 - 2b. Laudamus Te. with Sylvia McNair and the English Baroque Soloists. Female vocals.
Beethoven's Ninth Symphony, 4th movement. - Male Vocals and "Wow factor".
Verdi's Requiem - Dies Irae with Robert Shaw and the Atlanta Symphony Orchestra & Chorus and Tchaikovsky's 1812 Overture with Leonard Bernstein and the NYP. Bass drum impact. The bass drum is getting slammed pretty hard in both these pieces and if it just sounds like another tone without any sense of impact, it's pretty close to a deal breaker for me, especially in the latter where they stand in for cannon blasts.
I also usually test the fourth movements of Tchaikovsky's 4th and Dvorak's 9th symphonies. I'm not listening for anything in particular, but they're both among my favorite pieces, and if a system can't make them really jump out at me, I want no part of it.
Now on the non-classical stuff. This changes around quite a bit more and is much more subject to whatever I'm listening to at the moment. There are some things I'll hit up pretty reliably though. I don't take this as seriously and I don't generally examine them for things in particular as much as I do the classical stuff. I'm not as much of a classic rock junky as this will make me look, but they're known quantities that I know are of good quality.
Led Zeppelin - Bron-Y-Aur Stomp.
Gogol Bordello - Mishto. Builds by adding instruments one by one and is good for evaluating cymbal response.
Jethro Tull - Skating Away on the Thin Ice of a New Day
Pink Floyd - Wish You Were Here end of Have a Cigar / beginning of Wish You Were Here. I really like the way the sound withdraws suddenly, then seems to move around behind me sounding like a tinny little radio speaker and the contrast with the guitar at the beginning of Wish you Were Here