I can't respond intelligently on that... as I really haven't looked @ the underlying code.
But I would imaging, the formats define how much information is stored about each pixel, thereby giving you more flexibility on what you can achieve. This usually translates to increased memory pressure, and transform complexity (decreasing overall speed). This of course is highly dependent on your average use case image size.
But that's probably all academic anyway... if you need to use it.. you need to use it
As far as which one to use.... I'd suggest using a Singleton / or static final dec that copies the Image format configuration from the api, so you can change it in one place, and change that format wholesale if you decide that you need to change to another format that supports alpha per-pixel.