> (Plausible attack on a weak remote web service. Yes, cracking a stolen hash is faster, but it's not what the average user should worry about.)
This is almost a sound assumption for most web services[1]. However, this is Bitcoin. The only thing the attacker has is your hash. And you're using a payments system which economically incentivizes the creation of ever-larger systems for brute-forcing hashes. The network's hash power as a whole is estimated to be around 331 exahashes per second, so 68 bits of entropy would take one second to crack.
Correct horse battery staple would be cracked in fractions of a second by the full network. Eight common words would take 12 days. If we go further to 12 words, then we do get reasonable levels of security, but I'm assuming hashrates stay constant forever which is a bad assumption. And 12 word passphrases will already be about as much of a pain to remember as the 'password policy compliant' passwords xkcd was railing against.
[1] The most likely attack is actually credential-stuffing, not brute-force. xkcd is assuming you already use separate passwords.
Related: The password hashes for the xkcd forums actually did leak and it turned out most people's passwords were "correct horse battery staple".
No, not four random words. I mean the literal text "correct horse battery staple".