The probability of the foal being worse is actually more like 98.5% in your example. Because the actual random horse the parents are averaged with is not distributed uniformely, rather it is (at least for speed and jump) a sum of 3 uniform numbers, which is way closer to a normal distribution. Furthermore, averaging this gaussian with the parents slightly shifts its central value (average) towards the parents, aswell as lowers its standard deviation (foals are more tightly packed near this new average than they are near the general average in the wild, allowing for less variation).
Overall, while this allows getting better horses through selective breeding, the average amount of generations required is so insane it can just be considered broken (and it's easier to tame wild horse to get a better one, at least while it is possible). The only useful thing about breeding is, with "good" parents you get a better chance at "decent" babies than in the wild (but lose the possibility of getting super horses), so it can be seen as useful if you want to easily mass produce decent horses (as backup replacements for your mount, to gift decent horses with ease...). Obviously, all of this becomes worse when you start considering multiple stats at once.
Tweaking the bred horses gaussian to be closer to the parents and/or wider could make things more enjoyable (this is equivalent to adjusting the weights in the average).
Zboub
The probability of the foal being worse is actually more like 98.5% in your example.
Because the actual random horse the parents are averaged with is not distributed uniformely, rather it is (at least for speed and jump) a sum of 3 uniform numbers, which is way closer to a normal distribution. Furthermore, averaging this gaussian with the parents slightly shifts its central value (average) towards the parents, aswell as lowers its standard deviation (foals are more tightly packed near this new average than they are near the general average in the wild, allowing for less variation).
Overall, while this allows getting better horses through selective breeding, the average amount of generations required is so insane it can just be considered broken (and it's easier to tame wild horse to get a better one, at least while it is possible).
The only useful thing about breeding is, with "good" parents you get a better chance at "decent" babies than in the wild (but lose the possibility of getting super horses), so it can be seen as useful if you want to easily mass produce decent horses (as backup replacements for your mount, to gift decent horses with ease...).
Obviously, all of this becomes worse when you start considering multiple stats at once.
Tweaking the bred horses gaussian to be closer to the parents and/or wider could make things more enjoyable (this is equivalent to adjusting the weights in the average).