I agree with OzExpat about human nature, but then human nature is to a great extent what CRM is all about. In my CRM training I have never even considered the idea of modifying human nature, that's not the purpose of CRM and I have never heard any of my colleagues express such ideas. What we want to achieve in CRM training is an understanding of how we humans function in basically the same way as we understand the aircraft we fly and their strengths and limitations.
We can, however, modify viewpoints to some extent. Being in control is one such viewpoint as "being in control" does not automatically mean holding the controls.
I don't think there really is two sides to this discussion, only a big jumble of aspects to consider. There can probably never be a clear cut and simple solution to this, or any other situation, that will work every time. The NTSB report that b55 quotes also mentions that over half of the PNF co-pilots involved in the 37 accidents had less then 1 years experience.
It has to be a judgment call left to the captain to make in each situation, the important consideration being that the captain should choose the method that minimises his own workload and that situational awareness is maintained. A very fresh co-pilot, or a co-pilot the captain does not trust, will increase the demands on the captain.
At the same time the captain will want as much performance as possible from the co-pilot (and the rest of the crew) which he will not get unless they have some self confidence.
It's always easier to describe things using a scenario so here goes...
Let's say we have a take-off in a medium or heavy jet. The co-pilot is PF and he has a couple of years experience but the captain has only flown with him once before.
Take-off is at night and at 1500 feet they are IMC.
At 2500 feet, with the aircraft cleaned up and climbing nicely one engine fails.
If the captains first action is "Engine failure no.1! My controls!" he is basically saying to his co-pilot "Step aside little boy, you're not good enough to handle this" and in the process he is reducing his co-pilot to "a little boy". Whether the co-pilots pride is hurt is irrelevant as hurt pride has no place in this cockpit. However, the captain is going to need his assistant but he has probably reduced his assistants self confidence to "a little boy" by declaring him incompetent thus reducing the flight crew from two to one and a half. He might even have steepened the authority gradient to the point where the co-pilot becomes a passenger.
If the captains actions are to call the engine failure and to start the appropriate actions while leaving the co-pilot to do the heading/altitude/airspeed stuff he will bolster the co-pilots confidence and most likely get better perfomance from him but he also needs to keep half an eye on what the co-pilot is doing with the aircraft. If the captain takes the controls he will still have to monitor the co-pilot in the engine failure actions and possible attempts at relighting it.
If the co-pilots flies fine during the first few seconds he is likely to continue doing so but if the co-pilot is going to lose his marbles the captain needs to catch this before it happens. This can be easier than one might think; ask the co-pilot questions once every few seconds.
"How does the aircraft feel?"
"What is our airspeed?"
"What is our heading?"
If the answers get tight lipped or if he stops answering he is about to lose it and the captain should definately take over.
If he appears to be fixated on one instrument or initiates a turn in the wrong direction before turning the correct way he may be about to lose it.
If we add other difficulties to the scenario such as weather, terrain or performance the situation might be better solved by the captain doing the actual flying but he should then consider letting the co-pilot know why he is taking control.
Eastern 401 (Everglades) and a bunch of other tragedies have shown us the importance of always having one pilot flying the aircraft and maintaining situational awareness. When things start to go wrong this becomes even more important.
Again... there is no way anyone can say what is the best way, only the captain can say in any given situation and in my reasoning there is no such thing as "pilot error". The pilots always make the correct decisions in all situations considering the knowledge and information they have and how they percieve the situation.