MySQL Cluster High Read/Write Volume

We are currently investigating the possibility of using MySQL Cluster or another high performance solution to handle a large amount of write load. Seeing as all of these systems have specialized benchmarks juiced by their marketing department, I wanted to get a realistic view of what to expect.

I would first like to describe the read/write load we have and what I expect to see if what we want is even feasible.

We have approximately 150,000 INSERTS per second, and 155,000 UPDATES per second where 150,000 of them are increments (eg. UPDATE foo SET bar=bar+1) and approx 50,000 SELECTS per second. The rows are pretty wide, 150 columns or so, but almost all of them are BIGINTs. The data doesn't need to be highly durable, meaning that not every write needs to make it to disk right away. We can handle even up to a 10 second fsync. The cluster itself, however, needs to be highly available, so multiple copies of the data is preferred.

We are running in Amazon, so up to the largest machine they offer is what we plan on running.

My first question is, how feasible is this? According to their marketing info, they got up to 1mm requests per second on 8 commodity nodes. Does this mean I can achieve my requirements on 4? What tradeoffs are made for the performance characteristics they have achieved? How easy/difficult is it to manage as compared to a regular MySQL setup?

I appreciate any feedback, thanks.