Okay, found a little bit more info. From looking at the code in IDA Debugger, shininess doesn't mimic how other model commands are processed. When NWN is reading in an ASCII model and hits a command like ambient or rotatetexture, it will call a subroutine to filter the rest of the line of text to gather the data it's expecting. So if it's ambient, it's going to call a subroutine which checks for three floats worth of data. For rotatetexture, it'll call a subroutine which scans for boolean data like "0", "1", "true" or "false". Shininess is the only command that doesn't call a subroutine and, instead, looks like this. I.....dunno. I mean, I'm not sure if that's a coder "short circuiting" it to make it inoperable or maybe just building in the search function into the string search for the command.
FWIW, the way shininess is done is the same in nwmain.exe, processmodels.exe (the leaked internal Bioware compiler) but, interestingly, not in the standalone NWN model viewer.
Tchos: A week or two ago I came across some guy's basic custom content guide for NWN2 and everything was much more clearly laid out. I know the alpha channel can be used to reflect an environment map or transparency but in a situation where you have an RGBA and R, G, and B are black but the alpha has a grayscale in it (and you have no environment map), the alpha will pick up light. Like this. I think that's how the model was set up, anyway. Out of curiosity, does that look like what you'd think of when you think of a specular map? See, it's like so much crap is rolled into the same thing with the way NWN does lighting that it can be kind of difficult to pick something apart and say "Am I looking at something new or just something old but displayed in an unusual way?"