The way how I create my mpf_class effects the result after a arithmetic operation

[Torbjörn Granlund]

You need to understand the nature of binary floating-point better.

In short, the numbers you use, 6.4 and 3.4, cannot be stored exactly in
binary FP.  3.4 for example, is 1101100110011001100110011001100110...
where the repeting tail's length depends on the format's precision.

Your 3.4 stored in a IEEE double and stored in a default precision mpf
number will very likely be different, and in both cases relatively poor
approximations.

-- 
Torbjörn
Please encrypt, key id 0xC8601622