They essentially built an ad-hoc database complete with sequential keys and indexing. I'm curious to see what the resulting performance would be if they used a database that's appropriate for the type of data they have. I'm not familiar with non-relational databases but I suspect that a key-value or document store would work very well.