FHE is far too slow for any practical use. In a few more years, things might be different, but for now anyone marketing a practical FHE solution is probably lying.
I'm still trying to find performance numbers that prove it impractical. Otherwise, it just sounds like a problem that could be mitigated by clustering.
Here are some results from a research team that has been on the forefront of FHE implementations; note that this has been improved on significantly since last August, but you are still looking at minutes of computation for relatively small functions:
Don't forget Amdahl's law and Gustafson's law on the limits of parallelizing when the problem isn't P-complete. Either way, I disagree with your conclusion that it is misguided.
