FPGA can change to do the new algo in hours.
Until some day that one of those altcoin programmers discovers that his CPU has some interesting instructions like: FILD, FIST, FDIV, FSINCOS, etc.
Then the hours become months, and even after that the FPGA will have hard time beating even a cheap Atom CPU.
I tried to research exactly emulating 80x87 couple of years ago. There was exactly nothing available open source, with exception of an exact re-implementation of FDIV in Mathematica. The available closed-source code was not only expensive but had mandatory royalties and defense-grade NDA requirements.
Are you so sure about that? The floating point performance of modern FPGAs per Watt is much better than GPUs. Even in the 28nm Virtex 7 days TFLOPs were roughly on-par, its neck and neck now and the next gen FPGAs are leading ahead on the AI/ half precisions stuff. That Floating point performance gap was true several years ago but has rapidly closed since.
The types of instructions youre listing also take many many clock cycles on GPUs and CPUs, and can almost always be implemented faster in FPGAs