If the network latency is negligible, this sounds reasonable. However, the mapping of 2x IOPs to 2x cost seems tenuous to me.

>Someone who wanted an interesting research project might be able to pull back the covers on DynamoDB a bit by issuing a large number of requests, very carefully measuring the response latency curves for consistent and inconsistent reads, and looking at what model fits them best.

This is possibly interesting, but why shouldn't Amazon just provide this data? The data would useful to developers who otherwise have to guess.

And finally, why not allow users to place this cost on the write path instead? "W=3" seems like a reasonable option (given that consistent reads may be unavailable anyway).