It's not that the simplest methods are always the best on real data. Rather, usefulness and hardness are uncorrelated and you may be lucky.