There's no work QM is saving: it doesn't delay work or get to do less until when you physically look at something and then get to skip work for things you don't look at.
* Things that are in superposition would require more computation for all their many alternate versions (most of which aren't meaningfully interacted with), not less, and the extra work for managing superpositions continues forever if MWI is right. All particles are constantly entering new superpositions, not just individual particles in fancy lab experiments. Only the very simplest of particle interactions (like two individual hydrogen atoms interacting) are feasible to simulate with full quantum mechanics on classical computers.
* "Observation" happens through any physical interaction at all, including particle collisions. Outside of some individual particles inside carefully controlled experiments, most particles end up transitively interacting with most other particles near them.
If the universe was at all optimized for simulation costs for human experience, we probably wouldn't expect there to be be trillions of galaxies with hundreds of millions of stars each, for the smallest particles to be on the scale of billion-billionths compared to humans, or for QM to work anything like it does.
That time-steps for atomic interactions need to be picoseconds or smaller, but interesting biological reactions take minutes+ [eg,protein folding] is also a PITA.
QM-as-RNG for a simulated universe has been one of my favorite "for fun" pet theories over the years. Finding whether the electron went through slit A or slit B during the double-slit interference tests just feels a bit too much like running a Monte Carlo sim to me. I know it's just my human brain really trying to find an explanation that makes rational sense but it's a fun thought experiment.
I mean, quantum fluctuations during inflationary expansion are (in my understanding) what's responsible for the tiny differences in mass distribution that led to the eventual formation of gas clouds/stars/galaxies. The negative gravity during expansion worked to keep the matter in an extremely low-entropy (highly ordered) state, but those unfathomably small quantum fluctuations were blown up by the same unfathomably massive proportions as everything else.
Quantum computers are asymptotically more powerful than classical computers. This means that our world is actually harder to simulate (rather than easier as you seem to suggest) than a classical world would be.