It's easy to do this with a Lua script, but I doubt it will actually make a difference in recording time. Most of the power is drawn by the sensor in LiveView, not by the dimmed LCD.
Some numbers from a quick test (my battery is probably old, used and abused):
- Photo LiveView (30fps), no Global Draw, LCD at maximum brightness: 65%/h
- Photo LiveView (30fps), no Global Draw, LCD at minimum brightness: 61%/h
- Playback, LCD at maximum brightness: 30%/h
- Playback, LCD at minimum brightness: 24%/h
- Paused LV (similar to playback), LCD off: 22%/h
Feel free to do this test in a more controlled way (e.g. randomly switch between these 5 states, leave the camera in one state for say 5 minutes, and write down all the measurements; repeat until you fill a page with numbers.) You'll probably get solid figures about power usage in different modes (unlike my ad-hoc test), but I doubt they will change the conclusion.
You could try applying the tricks used by raw recording modules to reduce the CPU usage, but this needs a custom build.
Also, ML prevents powersaving at CPU level (WFI on 5D3), to keep things simple, but the difference is significant only when running with the main display off (see
FAQ).
Of course, all this means there is some room for improvement regarding power usage - it just needs some low-level research. If you are familiar with this area, feel free to jump in.