Why do you say only financial?
Arbitrary precision is good - but why should it be done in a way that is not natural to how machines work? But I admit, I have not looked at the code how it is implemented. Maybe it is done well (only saving integer part and an exponent). I only recall the BCD standard used in Turbo Pascal - which was not so great - and really mostly for financial use.
But there is no point in having fractions in decimal for computation, unless the result has to have a good representation in a specific base. Something you can send to a tax office. The decimal package solves problems specific to financial. And, yes, it can be used for other applications lacking other arbitrary precision classes in Python standard class library - but it would not be my first choice for other applications.
Hence I propose to have a different arbitrary precision package for applications that do not require 'decimal' precision. As for class hierarchy it likely should be a base class for the decimal package. (Then you could even add other bases as well.)