1. It could be cheaper using a texture format without alpha. You should look how many bits’ a format needs per texture element (texel). Fewer bits are in general more cache and GPU friendly. As some formats replace the missing alpha with unused bits you would not win anything at all.
2. The file type can only have an impact on loading time. As stated the best format is that one with as less bit’s as possible that is supported and still looks good enough for you. In general you should try using one of the DXT formats.
3. From the file formats that D3DX supports: yes. But you can always write your own loader that supports your own formats.
4. Sure. The DDS file format supports texture formats without alpha channel.
5. DXT1 works well for mask textures. Grey scales could be stored as DXT2 or DXT3 if the limited precession is enough. Otherwise D3DFMT_A8 should be a good solution.
6. Render target formats depends on your needs. In general D3DFMT_A8R8G8B8 is often a good choice. But as always it depend on what you render to this target.
As depth buffer format most today games use D3DFMT_D24S8 or D3DFMT_D24X8.
When it comes to format decisions you need always keep the following in mind:
1. Try to use a format that need as less bit per texel as possible.
2. Check if the hardware supports the format for that use case. If not have a fallback chain with other formats.
The ConfigSystem sample includes a table that can give you an overview of the formats that current hardware supports. This should help you to define your formats and fallback chains.