I agree that images typically are larger than code. I don't see why models are relevant when discussing the size of the SU code, though. They aren't distributed as part of the package.
To introduce a sample fact into the discussion, in a recent extension that John McClenahan and I wrote we provided translations to all eight of the non-Asian languages that SU supports (neither of us has even a smattering of Asian language ability and nobody offered to contribute translations). The total body of the Ruby code is 92KB, and the translated language support totals 356KB! Admittedly we did not try to optimize the implementation and the translation data could be shrunk considerably by various techniques. But even unoptimized, no single language's data was more than about half the size of the code.
I offer this concrete fact to underscore that without actually determining what part of the SU package consists of language strings, we don't have a basis to say whether including all translations would bloat the package. That's why I used the phrases "I wonder" and "unless".