That's how we improve fastest. But generally, the way we get results shouldn't result in a lot of stuff actually missing, which may speak to some bugs in there.
Which sucks, because I hands-down full-on no-buts prefer DDG's interface to Google.
I hope I'm not being uncool here, but a while back I privately suggested mining the data from "!g" queries as a way to see where DDG wasn't returning the results that people expected. Is there a reason you can't do that? In my case at least, I usually start with DDG, and if I don't get what I'm expecting, then I use a "!g" -- so you'd have a nice little database of results to perform comparisons from.
I'm going to switch again and start sending DDG results that aren't helpful.
Not sure how you could provide for that, though.
If you search for some things (try "stack overflow" or "red bull"), you seem to get a near-instant results for the thing judged most likely/relevant in a red box at the top of the page, then the "ordinary" results are loaded separately, I think in a separate javascript call.
They seem to have a some split architecture that keeps a few bits spooled up for quick retrieval, and say to heck with the rest. My first instinct would be that they're RAM-poor, and also wonder if, not being able to cache sufficiently in RAM, they're also being hamstrung by I/O performance on EC2.