It looks cool -- the graphs are a little strange though, for aggregates like "average response time" you'd prefer to see a different graph, preferably bar or box plot. If you keep a larger number of quantiles you can do really cool effects here like a blended-color stack plot for latency, something like this:
If you pick the color maps with narrow saturation peaks around the 50% and 95% level then you can get a really nice at-a-glance understanding of latency behaviour this way.
Re: the storage backend, not sure why you would choose acid-state here, as this basically rules it out for serious production use. The reason Doug picked zoom-cache in the first place is that it's designed with this format in mind: you have timeseries data that you would like to:
a) stream to disk
b) summarize and index this data in a way (successive power-of-2 aggregation/precomputation) that allows you to read an arbitrary time range with a constant or log-factor amount of disk (and RAM) I/O.
A service doing 10qps (pretty meager these days), assuming an acid-state overhead of 256 bytes per request (not unreasonable?), is going to generate 211MB of log data every day. It's not practical to store this on the heap.
G