I’d say the key insight with quantum computing is that its algorithms are about choreographing interference patterns among qubits such that wrong answers cancel each other out but right answers reinforce one another. It’s not just a matter of trying possibilities in parallel or “running different probabilities simultaneously” - the qubits’ states are complex combinations of 0 and 1 states, and they interact with and change one another. Simulating those interactions on a classical computer requires exponentially growing amounts of memory space and time as the quantum computation gets bigger. Trying to divide-and-conquer this simulation over multiple classical computers runs into the need for different parts of the circuit to know about each others’ state, limiting how much work can be sectioned off to be done by each computer in the group.