As a standalone statistic, 10% isn't very useful, because it's not 10% across the board for everyone. In some ways, it's less impressive than that, and in others, it's much more impressive.
The situation being addressed here is that certain graphics card drivers are notoriously buggy, such that processes that use normal accelerated graphics APIs will randomly crash for certain OS/driver/chipset combinations. Historically, Firefox has had to play whack-a-mole by finding patterns in reported crash data that says, for example, "ATI graphics driver x.y.z, with chipset Foo, under Windows 8, is showing an unusual number of graphics-related crashes, so don't use graphics acceleration on those machines." This results in slower rendering for those users in general; and, for those troublesome combinations that have not yet been blacklisted, you end up with users who see Firefox crash a lot (see, e.g., drinkypoo's comment below).
If you're not one of the people with a magically horrible combination of graphics card, graphics driver, and operating system, then this will make absolutely no difference for you. But for those poor users who have found this sweet spot of graphics card misery, performance will improve immensely (for those on the blacklist) and crash rates will plummet (for those who are not). And these users crash *so* *often* that just providing this workaround for their bad graphics card drivers will make *overall* Firefox crash rates go down 10%.
Hard data on *early* experimentation here the final numbers look even better: https://ashughes.com/?p=374