Note that your comment assumes that the 16-bit representation is a linear function of brightness, that is, value such as 12223 will be exactly 12223 times as bright as 1. This would not be true for your typical 8 bits per component pixel formats, although it might be true for 16 bits per component pixel format. (Is it? I don't know.)
This article does not mention gamma at all, but as far as I understand, the 32-bit floating point per component operates in linear light mode. This is imho a huge improvement.