A friend explained this concept to me as follows: let's say you're playing the Empire Strikes back and you're trying to escape from Hoth in the Millenium Falcon. You bust drop down the blaster turret from the bottom of the old rust bucket and start firing. You roll a really really good result. What happens:
d20: Crit! You kill a storm trooper.
d100: 100! You kill the entire storm trooper team trying to set up the e-web.
d1000 with sustain dice: 1000+ max sustain. Wow! You hit a critical structural element and collapse the entire complex (except the hangar you're flying out of) not only killing all the storm troopers in the base but also squashing Darth Vader.
If you don't want to have the potential for that kind of one in a million result, you don't need that much granularity. For his group, the d20 amount of granularity was as much as he wanted to allow to ride on a single die roll.
Here's the other element to this: if the game system includes about the same number of rolls as D&D, modifiers under +5% are worthless and even modifiers as high as 10% are only of marginal value. If 95% of the time, it won't make a difference whether you have that +5% or not, how often do you think it will matter that you get +2% or +1%? Pretty much never. (That's why in 4th edition D&D, while people get excited about a feat that gives +1 to hit, skill focus can give +3 to checks and very few people care. You roll enough attack rolls that the +1 will matter once or twice per session. You generally don't roll enough checks with any particular skill for the +3 to matter even once). In short, the added granularity of a % system is wasted. Unless you make lots and lots of rolls to accomplish all your tasks, it simply won't matter.