Is 128-bit "long-float" useful?
I guess the advantage of long floats is not only that they can span huge ranges, which may or may not be useful, they probably also have a much larger mantissa (I refuse to use the word "significand" for this) than a double, which gives your numbers a higher precision.
But as someone said, scientists love those types. Probably for the above reason. Note that the libraries are often called arbitrary precision libraries.
Scientists use this kind of stuff - and occasionally arbitrarily sized integers/floats/decimals.
For you, 32-bit or 64-bit is usually enough.
See also:
http://en.wikipedia.org/wiki/Arbitrary-precision_arithmetic