The recent revision of the IEEE Standard 754 for binary floating-point
arithmetic added specifications for decimal floating-point types,
specifically designed for use in financial applications. In the long
term, OCaml support for these may be a good idea.

Intel provides a library with a fairly portable software implementation
of the new decimal types here:

http://software.intel.com/en-us/articles/intel-decimal-floating-point-math-library/

And if you want to find out more about decimal arithmetic, there are
several talks and a special session on the subject at this year's
Computer Arithmetic conference, the early registration deadline for
which is tomorrow :-)

http://www.ac.usc.es/arith19/

John.