But "more hassle" by an order of magnitude that gradualy reduces when growing the number of instances.

But the cost keeps growing, especially when you need to go from 50 servers to 100

The OP only doubled their performance by rewriting in Rust though...

And depending on the amount of code, that's an awesome trade-off if OP can just have one instance instead of managing a cluster.

From what OP has said, there's probably a good change they can get more performance with some optimization.

You still need redundancy, so you’re operating multiple instances either way (or you’re playing fast and loose).