That's another approach, but gets into annoying subjective judgements really easily. I'm confident it's harder to get volunteers to do "work out how much credit this person should get for this change, as a proportion of total value of the page" (better than just wordcount) than "you're already doing edit review to keep out vandals? also select from these edit types to categorize it" at scale with minimal oversight. Long term you could even do some machine learning or basic pattern recognition to work out what type of edit it is automatically most of the time so the reviewer just needs to confirm.
A hybrid approach where editing or creating high-value (by likes) pages gets more credit, by applying some scaling factor to the karma payoff, is plausible? Not thought that one through in detail though.