I have to wonder if #3 isn't a quick thinking/pattern match for people with a CS background. Instant trigger/right for me. I'd think that people who work algorithmically should do better than a control group.
I'm defensive with that statement because there's quite a few "people studying the subject do no better than the average" experiments (in economic) but I think there's a difference between academic knowledge and applied knowledge so if you think about complexity problems often you should do better on this one.
The interesting takeaway for me is that environment probably matters. If I take a test like this in a very relaxed environment I'm probably slowing down automatically.
All three of them were a pattern match for me. The first one was an obvious "Subtract the difference, split the remainder" problem. The second one I immediately thought "What's the rate?", did a little math, and figured out that in both cases one machine builds one widget, so it'll still be 5 min/widget. The third one is a classic exponential growth problem.
Probably yes, but the point is that if you've seen similar problems before, the fact that they are counterintuitive trick questions is part of your pattern-match, and so you jump to a problem-solving strategy that manipulates the data you're provided in a way that's accurate rather than one that's not.
It's the same way you build expertise in any domain: proceed with your initial hypothesis, but then once experience proves you wrong, build a new, more accurate mental model. That new model will give right answers in a greater percentage of cases, but inevitably it's wrong too, and so you repeat the process.
The interesting takeaway for me is that environment probably matters. If I take a test like this in a very relaxed environment I'm probably slowing down automatically.