I'm sure someone has done quantitative analysis on tx encoding/types and you get a better guess at the avg from there.

k00b

I thought about that myself, but it really would not work as signatures have different sizes depending on time(as different encodings were used) and TX type.


raw_avocado

bitcoin

I want to say signatures take up X% of the blockchain. Who can help me find X?

This might be kind of naive, but you could probably get a close approximation by counting how many inputs there are and multiplying it by the avg size of a signature. 

I don't know enough about txs to say how accurate this would be.