Timing is a factor, too. For such a thing to be successful, you also need a OS/hardware combo that can draw the notification without slowing down using the main window (rules out early Mac OS) and users who think the added distraction of notifications is worth it. I would that added distraction goes down with screen size. I doubt it would have been a success on 640×480 displays, for example.
Also, Apple had something similar in 1997 or 1999, with Mac OS 8 or 9.
Obviously then it was then extended for things like a long process completing, etc.
But back in 1990 or 1995, there's wasn't a need for something like Growl. Your e-mail inbox and the occasional beep and modal dialog did the job just fine.
To counter your "most things are obvious in retrospect" philosophy, you might be interested in the "multiple discovery" viewpoint [1] which says precisely the opposite -- that, extended to design, essentially says that the need for a solution becomes obvious to people at about the same time, and that people will solve it in similar ways because they're facing the same constraints.
I also don't think multiple discovery is much of a counter to "most things are obvious in retrospect". Yes, design problems can get solved in similar ways by different people. But obvious-in-retrospect thinking is a cognitive error, where we presume that what's obvious to us is obvious to people in different times and conditions.
I was using NeXT OS, which is what later became OS X, circa 1990. Growl apparently launched in 2003. So that's a minimum of 13 years that something "obvious" was missed. And it's probably more fair to count from the mid-1980s, when GUIs first started becoming popular. That doesn't sound obvious to me.
Most things are obvious in retrospect. But it's a mistake to confuse your after-the-fact perspective for what was going on at the time. (For those who are interested, Dekker's "Field Guide to Understanding 'Human Error'" is a great look at how subtle and dangerous that confusion can be.)