Hold up. I'd like to hear from Wuille (one of the creators of segwit) about the size difference between a standard 2-input, 2-output transaction and its equivalent using segwit, for a fully-validating node. No need to attack with sarcasm.
BTW, I am also curious if the O(n^2) sigops issue can be solved in a much more simple way.
That leads me to another question I've been having that hasn't been answered as far as I know. If segregating the signatures out of the tx leads to a stable txid (malleability fixed), then why can't we simply fix malleability independantly by simply
ingoring the signatures when hashing the txid?