For some if it gets the right answer, time is unimportant - for others speed is paramount.
You should search through the archives for at least the following to see what a minefield it is!
bigint Big integer library by Richard Peters
mp_int by Kevin Sopp
Introduction of numeric_adaptor - switching the underlying mechanism.
Arbitrary precision arithmetic
[mp_math] a multi precision signed integer library
Using NTL and MPFR
Many of the problems are also in testing and documentation to get something that is good enough to review, revise, and revise and get accepted during the GSoC period.
(...)
PS Of course there is also a need for floating point ;-) But this is an even bigger can of worms!
To add a relatively new one, and currently my favorite: http://www.ttmath.org/ttmath Having the BSD license, being template based and header only, supporting integer and floating point, its precision can be specified by template, this one would IMO fit into Boost. Regards, Barend