Uhm, what's the point of quoting papers from the 90's? 
It's when all the *cool stuff happened*.
1. For portability of representation of constants between architectures.
2. Low working memory requirements.
3. Fast to encode and decode.
4. Simple to implement.
5. The code for the encoding and decoding of integers is very small -- only a hand-full of bytes.
2. Low working memory requirements.
3. Fast to encode and decode.
4. Simple to implement.
5. The code for the encoding and decoding of integers is very small -- only a hand-full of bytes.
I still believe this is a case of extremely premature optimizations, but would answer anyway:
1. Endianness is trivial to deal with.
2. That is irrelevant to fixed/variable encoding.
3. Not doing it is even faster.
4. Not implementing it is even simpler.
5. The code for not doing it is exactly zero bytes.
I don't see it as an optimization, but an element in a data exchange format that we should consider if we want to support signing transactions on everything from a tiny 8051 based smart card to a huge rack-mount server and everything in-between. The nice thing about this scheme is that, while endianness may be "trivial to deal with," we don't have to deal with it at all.