Here's a corner case for you.
A user produces a small amount of confidential data and would like to delegate a third-party to perform some computation on it and return the results. Data and results are to be kept secret from the third-party, you can assume a "curious but honest" threat model.
My first naive intuition was that this would be an obvious application of FHE, but now I wonder... is there any way to tackle this use case using sMPC instead?
In other words: split the input data into a number of "workers" which are all controlled by the same third-party. Is there any sMPC scheme that is robust to such a scenario?