That's because at 100,000 years the technique pushes the limit of potassium-argon dating since the presence of potassium-40 is minuscule compared to a fossil billions of years old. The amount of K40 you quoted is also small, so depending on the presence of K40 in the sample a varying result would be produced. Older samples need less K40 present to produce accurate results whereas newer samples (100,000 years is the limit) need more.

The intention with these methods is not to show conclusive proof that the earth is x number of years old. Any method will give you a rough estimate as to the age of the sample. Sometimes that method will have a deviation of plus or minus 5,000 years and sometimes it will deviate plus or minus 50 million years. As these methods improve and as new methods for dating arise, gaps will close. Regardless, and what many people don't understand, is that a million years of time is incomprehensible to humans, but in earth history it's actually nothing, less than 1% of time. I agree, methods need to improve, but trying to discount them because there's some deviation in the results is absurd especially when it's the best we have in determining a blueprint of our world.