Ah ok! Perfect, thanks, that's what I was missing. Makes total sense now. In any lang (even in ruby!) you could have separate thread pools(or EM loops, whatever) for the s3 requests & web handlers. But because node only has one EL, and node interprocess comms is awkward, its tricky. Gotcha, cheers.

I wonder if using something like async.eachLimit would have helped; it might prevent the s3 batches from flooding the loop & give a chance for web reqs to interleave, but probably at a cost to the median resp time.