IMO any kind of automatic rounding and precision should not be part of a basic_decimal type at all!
I assume you mean that we don't need a decimal type with a fixed number of decimals. There is an extension to the propsed standard that allows fixed number of decimals so there seem to be some who need it. I know that I need fixed number of decimals in the software I am working with and that I often both have objects with round_to_nearest and round_toward_zero in the same application. If I had a decimal floating point type available I could probably have used it. It would make some things easier and other things more difficult.
The behaviour of basic_decimal should stick to the decimal proposal of Mike Cowlishaw and he explains very well in his FAQ section why it's not a good idea to have normalized decimal arithmetic:
Why do you mention "normalized decimal arithmetic"? Isn't that a way store floating point numbers or do you mean something else.