I’ve actually jumped in incorrectly with my previous post, mistaking it for another issue with height blended layers.
This one is not a problem at all.
So, what actually happens is something like:
When the weight of layer one is 1.0 and weight of layer two is 0.0, you first interpolate between layer zero and layer one using scalar value of 1.0 , and on the output you get only layer one.
Then, you interpolate between result and layer two using scalar value of 0.0 and on the output you get only previous result, which is layer one only.
When the weight of layer one is 0.0 and weight of layer two is 1.0, you first interpolate between layer zero and layer one using scalar value of 0.0 , and on the output you get only layer zero.
Then, you interpolate between result and layer two using scalar value of 1.0 and on the output you get only previous result, which is layer two only.
But when the weight of layer one is 0.5 and weight of layer two is 0.5, you first interpolate between layer zero and layer one using scalar value of 0.5 , and on the output you get 50% MIX of layer zero and layer one.
Then, you interpolate between result and layer two using scalar value of 0.5 and on the output you get MIX of previous result, with layer two, which is essential mix of all three layers, while technically, you have this spot fully covered by two layers.
To get desired blend, it might be an option to multiply layers by their weights( layer zero should be multiplied by one minus sum of other two layer weights) and add their results together.