"Ninety percent of everything is crap." -- Theodore Sturgeon
Ted Sturgeon was talking about writing when he came up with his law and I think he was correct in setting his cut point for excellence at this point. However, when you talk most things - and specifically things like software - I think that his law is overly critical.
Ten percent is probably the proper cut-point between excellence or genius or amazing or whatever other superlative adjective you want to use and everything else, and you probably need to meet that high bar in writing or any other artistic endeavor to be seen as original and making an impact. However, just because something isn't brilliant or original doesn't mean it isn't serviceable or useful - which is the bar we set in most engineering fields. As such, I propose an updated form of Sturgeons Law for engineering that I call the 10-60-30 Rule: 10% of everything is excellent, 60% of everything is useful and/or serviceable, and 30% of everything is crap.*
Original and brilliant need little explanation. However, there are many parts of our systems that don't require great genius - just good solid code that does what it's supposed to. This good code comprises 60% of most systems (and I would presume 60% of writing, art, and music creations). The final 30% is bad - it needs rework or excision and replacement to move it to the categories above.
One other aspect of this rule? The corollary that, even after refactoring and removal of the bad code, the ratios above are still descriptive to the body of code - it's just that you've raised the bar a bit. And this is why code maintenance is so important - it's hard to make the code better by improving either the good or excellent code. But by improving the bad code (which is easier to do), you've improved the quality of the whole code base, and (almost certainly) uncovered new issues in the good portions of the code that can now be improved.
So watch for that 30% of bad code. It's a high enough percentage that you almost certainly know where it's at. Then take the initiative to fix it. It won't get rid of bad code (remember, once that this bad code is fixed, another 30% becomes the worst in the code base), but it will raise the bar on your code's quality. And the people who find the code that you have left will be grateful.
*Actually, you can break down the larger categories further: 10% is excellent, 20% is really quite nice, 40% is good, solid work, 20% is bad but salvageable, and 10% reeks and should be replaced. However, "10-20-40-20-10 Rule" doesn't roll off the tongue as nicely as "10-60-30 Rule" does and, since what you do with the two classes of good and bad code are the same - leave alone and rework ruthlessly, respectively - the latter seems to be better. True aficionados of software numerology may have fun with the fuller definition.