I wish more ZK and FHE developers would present their overhead as a ratio (encrypted computation time vs. raw computation time) rather than just saying "we can perform N operations per second." This is more hardware-independent and gives a very indicative number: How much efficiency will I sacrifice if my application adopts a cryptographic approach instead of a trust-reliant one? It's also generally easier to estimate because, as a developer, I already know how much time my computation takes to compute the raw data, so I can simply multiply by that number. (Yes, I know this is difficult because the operations required for execution and proof are heterogeneous, especially in terms of SIMD/parallelization and memory access patterns, so even the ratio is somewhat hardware-dependent, but even so, I still think the overhead factor is a good number despite these flaws.)