I think arduino needs a 64bit double to be fully implemented.
It is often asked for, but have seen no code yet...
There are not so much sensors that give more significant digits than a float can handle. Note that 6 significant digits means a dynamic range of a factor 100000.
Why do you need doubles?