A few years ago there were papers on evaluating simple logic circuits in an FHE context and it took 2h hours for what was basicially 5-6 NOR gates.
Our team has been working on making FHE more accessible to engineers via a compiler; we've found usability to be a much bigger obstacle than performance.
You might be surprised to see how far performance has come! For (an admittedly small example of) matrix-vector multiplication, we can do key generation, encryption, computation, decryption, and compilation in less than 5 seconds on a MacBook [0].
As (maybe weak) evidence the progress on practical implementations of PCPs/SNARGs