I need load all documents in a big collection(millions), but only two fields.
My application is build in java. I don't satisfied with the speed, so I tried cpp and c driver. the result is nice when server is warmed up. But I found when the server is just started, the cpp driver is far slower than java's, the c driver is worse.

btw: it's a covered query, but without hint it will be a basic cursor. What confused me is when I add a hint, it will be slower despite indexOnly is true. @.@

The benchmark(approximate):

lang cold warm
java 48s 6s
cpp 200s 2s
c 400s 3s

I wonder how this happens?
Can you explain it? How to improve it?
Thanks