It should have no effect on performance.
SurfaceViews have two parts, the "view" part, and the "surface" part. The "view" part is a transparent hole that fits in with the other views, the "surface" part is a completely independent layer that is composited with the view layer by the system. The video is being sent to the "surface" part.
If you override SurfaceView's "view" renderer, you'll get a hardware-accelerated Canvas for a View that is normally completely transparent (so if you erase it, you better use an alpha of zero and the correct transfer mode).
If you attempt to render on the "surface" part, by getting a Canvas from lockCanvas()
, you will either fail (because the video effectively has it locked), or succeed and prevent video from being written to it.
The system compositor is going to have to blend the "view" and "surface" layers no matter what appears in the "view" layer, so making a few more pixels opaque isn't going to have a measurable impact.
Update: see the graphics architecture doc for more details on Surfaces and composition.