Double buffering and fancy drawing tactics are only half the story. The other half, some would argue the more critical half, is to limit how much of your control is invalidated.
In your comments, you mention that you use RedrawWindow(handle, @R, 0, rdw_Invalidate or rdw_Frame). What are you setting the R rectangle to? If you set it to your client area rect, then you are redrawing the entire client area of your control. When scrolling, only a small portion of your control needs to be redrawn – the slice at the “trailing edge” of the scroll direction. Windows will bitblit the rest of the client area screen to screen to move the existing pixels over in the scroll direction.
Also check whether you have set your window flags to require full redraw on scroll. I don’t recall the flag names offhand, but you want them turned off so that scroll actions only invalidate a slice of your client area. I believe this is the Windows default.
Even with hardware accelerated graphics, less work is faster than more work. Get your invalidate rects down to the absolute minimum and reduce the number of pixels you’re pushing across the system bus.