Someone on Talkchess did a test getting the top engines to play against each other in an opening which required an understanding of pawn breaks, such as the King's Indian or Stonewall. Houdini, Komodo, Rybka and Stockfish all struggled in these positions refusing to advance their pawns, and pursuing ephemeral tactical gains instead. At blitz time controls, engines overestimate the value of space and refuse to play obvious freeing moves.
Another area where engines fall down is planning for the endgame. Human grandmasters aim for certain positions which are known to be theoretically won or drawn, but engines are always caught up in short term goals, and quite willingly enter positions that are lost strategically. I think that it was Larry Kaufman who said that top engines' evaluation functions at around a 2300 or 2400 Elo level, but they never makes tactical mistakes.
Both Kasparov and Anand have commented on Hiarcs' positional understanding being more notable than other engines. Tord Romstad, one of the authors of Stockfish, acknowledged this in a fairly recent comment when he noted that on slow hardware, Hiarcs plays surprisingly well suggesting its strength lies more in its evaluation. Hiarcs has an elaborate evaluation schema, but the top engines tend to rely on a fast search.
Some engines are noted for playing in a sharp style: Tao, TheKing Tribute, Thinker and Zappa Mexico Dissident Aggressor. Some engines have parameters for sacrifices or king attack or their contempt works this way: Amyan, Fritz 10, Gambit Tiger, Junior 7 and Rybka Dynamic.