This issue was really screwing with me in CityEngine 2016.1. I'm glad I found this post.
I can verify that this issue persists in ArcMap 10.3.1 and that setting the 'NoData Value (optional)' setting to 0 when using Clip (raster) is the only way to solve it. It will set the 'Pixel Depth' to 8 Bit and the 'NoData Value' to 0, 0, 0.
Do not set the value to blank! Blank sets the 'Pixel Depth' to 16 Bit and the 'NoData Value' to 256, 256, 256.
You can see the difference in Windows Explorer.