Exactly what I'm asking. Anyone can make read benchmarks fast. He's adding index optimizations such as clustering that kill write performance. Balance please.
The original Digg entry specifically makes the entire point that Digg is willing to sacrifice write performance to improve read performance (which they did to a massive degree with Cassandra), and then demonstrated how absolutely horrendous their read performance was.
Need more be said? Seriously?
Cassandra is hugely biased towards making writes fast [1] and there's no indication in either Digg article that they (somehow) changed that fact in their deployment by tuning. Digg loads the complexity during the write by denormalizing but that load is on the application server not the data store.