When I wrote about Chekhov's Gun Syndrome, I had more than one idea in mind. My advice there was that not every situation needs some kind of dice check every time it is encountered: not every bridge should require Dex checks to cross without falling. Instead, a potentially dangerous situation could postpone any dice check until a specific action is taken: make a check to avoid falling when an opponent tries to make you fall, or when you try to move too fast.
But the related idea to this is, if there is a chance for something to go wrong, what should that chance be? Should the odds be adjusted based on real-world probabilities, or challenge levels?
My advice here is sort of the inverse of the advice above, or maybe its complement: you should not pick the odds based on real-world probabilities or scale the odds to match the perceived level of the challenge. Instead, you should make sure that interesting stuff has a decent chance of happening.
You may have noticed how much I use that situation roll (5+ on 1d6) for just about everything. It's not just because it's easy to remember (although that does figure highly) or because I want a unified system (screw that!) It's because a 1/3rd chance of things going wrong (or changing things in your favor) is about the right "balance" between things staying the same and things changing into something more interesting. The rope bridge might break? The odds should be 1 in 3. You might drop an item when surprised? Same odds. You want to bust open a door with one swift kick? 1 in 3 odds, otherwise you've just alerted anyone on the other side, although you can still bust down that door without any other rolls. You can modify these odds slightly (stronger rope makes breakage less likely, too much weight makes it more likely.) But unless a character does something that should obviously automatically succeed or fail, you don't want the odds to change that much.
Why? In a sense, it's because of Chekhov's Gun. When you add a dangerous bridge across a chasm, it's not very interesting if the bridge never has a chance of breaking, or if the chance is so ridiculously low that it's unlikely to happen. Why waste time rolling, if it's unlikely to happen? Why waste time calculating the real odds, when really you want things to go south more often?
I've mentioned before that I've played a lot of The Sims 1 and 2 in the past. There are a lot of situations in there (toilets clogging, showers breaking, burglars arriving to burgle your home) that don't happen all the time, but happen a lot more frequently than they would in real life. The game designers tweak the odds so that things go wrong often enough that the game isn't boring, but not so often that the game just becomes frustrating.
This should be your guideline when designing dungeons or running the game. Things have to have a chance of going wrong in unexpected ways, to prevent boredom (even your own.) But they shouldn't happen so often that it seems like your're being capriciously antagonistic. 50-50 is too random; 1 in 3 is a better balance. About the limit is a 1 in 20 chance; anything rarer than that is too rare to really matter.