I see it both ways....
On one hand, you need to throw in a few difficult questions into
the mix to separate out the good ones from the average....
On the other hand, if you were to set the questions to be too
difficult, you may end up killing everyone until you have to
moderate radically. Such a situation may not necessarily reflect
the aptitude of the students.
For example, in one paper I took a few years ago, there were
four long questions of 25 marks each. Three of them were
super-difficult questions which hardly anyone could do, myself
included.
In the end, I got A-, whereas a couple of my friends got C+.
Why? When they faced those three difficult questions, they wrote
virtually nothing in the answer script and got nothing in return.
When I faced those three questions, I wrote a whole bunch of
nonsense to try to analyze the questions and actually received
marks for that nonsense.
So, by setting those ultra-difficult questions, was it a true
test of aptitude? I don't think so. It was more like a test of how
much nonsense you can come up with, really. 
