>
> I can't remember who I saw discussing this idea. Might have been Vitalik
> Buterin?
>

Yes, he described it in an article a couple of months ago:

http://blog.ethereum.org/2014/01/15/slasher-a-punitive-proof-of-stake-algorithm/

but it is an old idea.
For example, I've mentioned punishment of this kind in discussion about
PPCoin when it was released in 2012, and, I think, it was described in
Etlase2's Decrit design.

Also, I and Iddo did some research on pure proof-of-stake, and it seems to
be feasible, in the sense that there are no obvious problems like "nothing
is actually at stake". (Unfortunately I can't refer to it now as it isn't
published yet.)