I have heard people say that scrypt is better than SHA256 as a basis for cryptocurrency because it doesn't put all the network power in specialist ASIC boxes, but then stuff like this happens.
Do you think Harvard would have paid for a custom-build scrypt-coin miner if someone wrote a real academic proposal for it?
I don't see why not, custom-built just means grabbing consumer grade gpus, with the "best" one currently the R9 270, which is ~$200 usd each. I don't think you need a huge budget so for academic reasons, why not. Plus, there are many scrypt coins out there (or they could even create harvardcoin) they could use for research purposes.
Don't need to necessarily use the mainstream profitable ones
From my understanding, scrypt ASICs have been held back by the cost of memory hardware as scrypt requires more memory usage. There has been talk of forking existing scrypt currencies to require even more memory to counter ASICs further but so far I think only a few new coins have done this.
Its memory requirement of 128KB is a compromise
between computation-hardness for the prover and verification efficiency for the verifier.
You don't want verification of a proof-of-work to take a lot of resources, since every client has to perform it.
Do you think Harvard would have paid for a custom-build scrypt-coin miner if someone wrote a real academic proposal for it?