Humans read base58 bitcoin addresses (20bytes) as ASCII numerical representations. Saving these into notepad is going to result in 34 bytes and will be compressible given a large enough sample. The block chain is a binary beast and is very efficient. Addresses are saved as 20bytes. If you can compress this (regardless of set size)  we are all in trouble.

All other data consists of arbitrary data (ie output values) and indistinguishable (hopefully) random data (cryptographic hashes and signatures).

I suppose if someone embedded structure into the block chain, ie through output values or non-working addresses (no private pair) you could somewhat compress it.