For what its worth, programs like FRAPs, Camtasia, etc normally dump the video near-raw with very little processing just meta data basically, to be later dealt with by another tool to create into a video file. This should be light on resources.

Xzins method attempts to accomplish this by pushing the data off to another computer and the data is just streamed there, lots packets are lost and not resent etc for the lowest overhead.

I use Camtasia, i just select one of the slave windows or even a YouTube sized subset of the UI window and use that. If you can avoid selecting a window that spans between multiple monitors and especially multiple video cards, or similar issues. If you are not going to want the sound in the end, don't capture that and remove it later do that upfront.