Evaluate/Optimize Unprimed Get Performance

[sdesai]

We spent some effort optimizing get assuming a response from cache (that is, we took some stabs at optimizing the get from cache algorithm). There may still be optimization worth going after in the case of gets which involve the network (collapsing/batching paths on the way out, merge results back into cache) which, even though the network cost is likely the long pole end-to-end, may result in less CPU contention for UIs.