[latchkey]

1) You are making an assumption that the original design decision was made around cost. I bet the more likely fact is that they made the change and found out that it cost them $160. Remember, that was when AppEngine was a _beta_ product and it seems they got lucky the first time.

2) They seem to have an extreme use case. No one is going to argue that maybe AppEngine doesn't fit the bill for them. Or, one could argue that doing 6.5 billion writes times a large number of customers, across multiple datacenters is something that a lot of databases would choke on.

3) Running more queries, while admittedly hacky is less expensive than doing more writes.