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.
