These computational chemistry problems make it easy-ish to eliminate candidate drugs that would be impossible in practice. Think of it as narrowing the search space, or reducing the size of the haystack rather than trying to find the cure/needle.
This would probably not work since bitcoin miners are all ASICs these days.
I feel like the question was just trying to see how much computation would help, rather than whether the chips are actually capable of doing it.
It's a reasonable question, but it's hard to evaluate the processing power of those chips because of how very specific those circuits are.
He said "was used". Bitcoin was run on CPUs and GPUs for a long time. Ethereum and other coins are still good on PCs i think.
It definitely would not work, they can do nothing except a partial SHA256 hash- not even a complete one.