I don't understand what the debate is here.

Is it this?

1. The supposed problem with brain wallet passcodes is that the ones that are easy to remember are insecure and the secure ones are hard to remember

2. The supposed problem with writing down the passcode is that then it is insecure.

3.  So supposedly what is required is some method of only having to remember some  simple piece of information that then enables the owner to retrieve the passcode without having to write it down anywhere.

If 1. and 2 really are a problem, then the solution is surely to use something which is un-guessable and un-brute-force-crackable to generate the private key. That is surely easy?  e.g. a jpg (then all you have to remember is "the photo of me with the clown hat" - though then you have to do your own hashing) or some unique sequence of text (e.g. "the first paragraph of the letter granny sent me on my 16th birthday"). Obviously don't lose the clown pic file or the letter from granny.  

But I think 1. and 2. are both dubious.

Re. 1. , if there are 10,000 English words, then choosing seven of them at random gives 1 * 10^28 combinations which should be enough for a few years yet even with terrahash asics.

Re. 2. a passcode is less obviously a private key than is a private key. So writing it down is less insecure than printing out and storing your private key (which is generally regarded as a very secure way to store BTC)

So IMHO the solutions are neat but they answer a problem that isn't really a problem.

Edit: changed 10^20 to 10^28