I think each r,g,b, vector is 8 bit total of 24 bit. With alpha another 8 bit = 32 bit total. I heard it's best to add any mask/grayscale map to your alpha which only adds one more 8 bit instead of making another 24 bit texture. Also not sure if this is correct but if you set the compression in photoshop to 8 bit grayscale…
on a side note this is quite weird, I'm not sure whats causing it but when adding the gradient with noise as a vertex color angled across a 100 by 100 segment plane you get an odd interference pattern when the octives are 8 or higher. And It only happens when angled.