I think it should be relative to content. As in versus actual text on the screen. This metric can also be applied "per-page" and "per-site", with less ambiguity for SPAs; every new load brings in more bytes, but also more text, thereby contributing to the ratio.

Even better - fixed site-wide assets (i.e css, js) to features; hence loading an entire framework only to use a small % of it's features is penalised.