If I am correct, this project is meaningfully extensible only if it is possible to communicate with exterior world other than tau-chain itself, such as www.
But the problem is, in verification processes of each blocks, the state of exterior world may vary from time to time.
If all the information is contained in tau-chain itself, the whole db size will be absurdly immense, because then all www is cloned to db.
Does pruning of blockchain or hashing of data help?

Also, Ohad describes agora as a market for developer. Do you intend to check if production code passes all the tests on non-turing-complete platform on tau-chain?
It sounds like developers can only write a code in non-turing-complete languages.
Perhaps, for example in C language, limiting while loops, function recursions and memory allocation size would make C emulatable on tau?
Maybe ethereum is useless in real world, but as far as I see, tau has the same inherent problem about turing-completeness.

I think this is one of the most eccentric project in the world now, but I actually doubt if it really meaningfully works.