It's probably not that there are weak testcases but that it's a badly set problem. You've got 3s TL with $$$10^5$$$ constraints that can be answered in very lightweight $$$O(N)$$$ — that's just asking for trouble. It's a common difficulty as a problemsetter to cleanly differentiate a heavy solution that has a better asymptotic complexity versus a very light solution with worse asymptotic complexity, while also keeping the time limit within reason — and it just seems the authors failed to do that here.

So yeah, the optimizations do help a lot, but if the problem was set with more care, it still wouldn't pass. The author in the editorial apologizes for not testing with pragmas, and it's a very old contest so fair enough — it's a more common trick nowadays (which I think shouldn't be allowed, but I digress)