Thanks for that article, this is an interesting new spin on merge mining! The difference between thermodynamic and economic security is something that comes up repeatedly in people's critic of merge mining (and in my case Namecoin). Armadillo is an interesting idea to increase security on shorter time frames (i.e. preventing double-spending attacks against an exchange like you describe in the article).
One thing I wonder about is this: With Armadillo, you need to place the commit-to-parents vector directly in the Bitcoin coinbase input, right? That means that unlike classical merge mining where you have a Merkle tree of mined chains, you can only merge-mine so many (a rather small number) of coins with Armadillo. I also think that miners were concerned about coinbase size in the past. Do you think that could be a problem?