It'd be more useful if the graph was in terms of comments per story per month, I think. As it is, the data could just mean that stories hit the front page, get discussed, and disappear increasingly quickly.
Stories hit the front page and disappear as a part of Hacker News's time weighting for determining ranking. It's extremely, extremely unusual for an article to be on the front page more than a day.
Right, but they can be on the front page for much less than a day, depending on how quickly we move onto discussing the next thing. And that "much less" can shrink further over time.
Why is it necessarily a bad thing that the number of comments has grown? This only matters if the overall quality of posts has decreased. More comments can also mean more good posts.
The number of comments has been growing significantly; perhaps change won't be bad.