i'm interested to know that is there any speed difference between c calling convention and stdcall?
and when we REALLY need to use c calling convention?

The obvious case is where the callee doesn't know the number of parameters pushed on the stack (ex. wsprintf, et al.). Speed is quite relative - both techniques can be optimized in different way. The most important aspect in terms of modern code is that the stack memory is already in the cache.

I've alway liked the ENTER instruction - quite powerful if used fully. Reserves and copies data across stack frames. Requires a lot of planning to make it work effectively though.

For anything worth calling, calling convention overhead should be pretty irrelevant - inline stuff where it matters.

(Of course every bit of extra speed doesn't hurt - but if you're looking to "reduce the overhead of calling PutPixel", you're Doing Things Wrong(TM)).