Personally problems like that aren't a big deal. They are hard, but man there are some stupid ones in the mix that get exceptionally high ratings.
I think that in general its a bad practice to take games between low rated players where in some random position the players missed a tactical continuation that is simple to find, but the BEST thread of the continuation is debatable. These problems often have the solver continue on long after the good move has been played, forcing the solver to guess at what more could possibly be gained in this already winning position. The difference between these elongated variations is often small; using the stockfish analysis might show a difference of 0.2 or so, sometimes its actually no difference.
Here is an example of what I mean:
http://en.lichess.org/training/50686After you find Qxa7 as the best move, there are lots of good responses to black's reply of Ke7. A human is not going to see a 0.1 stockfish difference. So, just using engine analysis shouldn't be a tactical training tool. I think there should be a minimum variation eval difference to qualify as a problem. For example saying there must be at least 1.5 difference between the best move and the alternate move.