For as long as I can remember, I have always preferred broad concepts over details. My trip to Schrodinger over the past 2 days has been filled with interesting insights, many of which of conceptual nature. I particularly enjoyed some teachable analogies that can serve us well when we explain computational chemistry related material to students. Art Bochevarov of Schrodinger provided a nice way of thinking about the meaning of basis sets and levels of theory in quantum mechanical calculations. I really enjoyed his analogy to cooking. I am sure you are all used to seeing things like “B3LYP/6-31G**” in papers that deal with computation. The B3LYP moniker stands for the level of theory, whereas the 6-31G** portion defines the basis set. The analogy to cooking is in comparing the basis set to ingredients you get at your grocery store (for a given meal, you might want to buy chicken, celery, pepper, salt, coriander). The level of theory is best compared to a cook as this is more or less about which recipe to follow and how to use the ingredients you have in the best possible way. Of course, the dish itself is the final product of a calculation. A lot of care must go into selecting proper ingredients (the basis set). Equally important is the cooking part. If you do things wrong, you might end up with a dish that will have chicken and duck in it, which is not what one wants to see. I think I will always remember this analogy. Among other things in Boston, there were some great insights into how to choose the right basis set for a particular “dish” one wants to cook.
It was interesting to note how many industry folks were at the workshop I attended (everyone but me was from industry). I was trying to see why and apparently there are some compelling reasons to go for high level calculations these days, particularly when it comes to molecular properties such as pKa, logP, and conformational properties. Below is a cool paper that proves the point. This work was mentioned by Art (he also offered some critique of the authors’ basis set/level of theory choices). According to Jaguar, compound 14 has an optimal torsion angle of 0o, which is suboptimal in terms of binding to the pivotal Leu398 of PAK4 kinase (the protein target in this case). With this in mind, a considerably more potent derivative 16 was designed, prepared, and validated. Because of a steric clash, this molecule features the torsion angle of 30o, which enables it to properly interact with Leu398. We’ve got to use more quantum mechanics, ladies and gentlemen.
amphoteros 
I don’t think this reasoning is correct. Normally I would expect some ortho substituent to tilt the aryl next to it out of plane. But in case of this pyrimidine 16 the NH of the methoxyethylamino should stabilize planar conformer with the NH binding to the pyrimidine N. (There is a weak clash of ortho H but they are enough apart to accommodate planar orientation. ) The preferred conformer of 14 is probably correct, being stabilized by the lone pair repulsion on nitrogens and the aryl-aryl conjugation.
There might be a number of other factors, I agree. I need to ask the authors about what their thoughts are. But the dihedral scan correctly picked up the winners. Let me think about what you mentioned.