In one of the Epic’s live training was said that translucent can be more efficient than masked on mobile devices, here is the link with timecode:
Is it really so? And if is what are the reason of that?
Not on all mobiles devices and not always. Typical cause for translucent being situationally faster is the fact that alpha test is done by dropping fragment in fragment function using a conditional and conditional depth writes are slow on most mobile hardware.
Translucent being faster than masked is subject to a lot of factors, but the most contributing is shading complexity and overdraw.
As material becomes more heavy and overdraw increases, it approaches the breakpoint where masked will be faster than translucent.
To sum it up, for simple forward shaded material, translucent will be faster than masked, provided that we are talking only of one layer of those materials on screen. When you step away from said assumptions, the statement that masked will be faster than translucent is not necessary true anymore.