Even if brain wallets are considered vulnerable, i think a long prhase like that should be secure enough to proceed with. The risk part is the process of making the wallet, remember to never use an online address generator, always generate them in your computer and if you can generate them under Linux that would be better.
I get your point and obviously a very complex passphrase can lead to a brainwallet that is infeasible to be cracked.
However, the point of a brainwallet is to be remembered, isn't it? So, if you use
duboli kufuzu naleqy lyfety dohihy rutoqa rageli kilyly vetyka himujy teluzi fuoda va as a passphrase, essentially it's much worse than using a CSPRNG to generate 128 bits of entropy to be used in a BIP39 wallet.
However, your result doesn't contain anything that is easy to read or write like "iqogu". It is not a word, but just random letters. So the user has to write them down one by one with a high chance of mistake.
In other words there is no difference between "sylui duboli kufuzu..." and "C85AFBACCF3E1EE40...." in how easy it is to read or write.
I totally agree with pooya87 on this one.
I would also add that BIP39 is a standard, which means it's properly tested and used.
Your method would be superior to the BIP39 standard, only in the sense that BIP39 is limited to a small dataset of 2048 words.
But again, unfortunately, the idea is not very useful.
