It's not a bug, it's an intended feature.
From the github page that you linked to:
How does blockparser deal with multi-sig transactions ?
--------------------------------------------------------
AFAIK, there are two types of multi-sig transactions:
1) Pay-to-script (which is in fact more general than multisig). This one is
easy, because it pays to a hash, which can readily be converted to an
address that starts with the character '3' instead of '1'
2) Naked multi-sig transactions. These are harder, because the output of
the transactions does not neatly map to a specific bitcoin address. I
think I have found a neat work-around: I compute:
hash160(M, N, sortedListOfAddresses)
which can now be properly mapped to a bitcoin address. To mark the fact
that this addres is neither a "pay to script" (type '3') nor a
"pay to pubkey or pubkeyhash" (type '1'), I prefix them with '4'
Note : this may be worthy of an actual BIP. If someone writes one,
I'll happily adjust the code.
Note : this trick is only a blockparser thing. This means that these
new address types starting with a '4' won't be recognized by other
bitcoin implementations (such as blockchain.info)