They're not biologically weighted; it's just probablilities. Punnett squares are just models for predicting inheritance. Like all models, they are indeed not 100% accurate, and the 25% weight is not true for all traits.

Some traits are more dominant than or can blend with other dominant traits. We see this a lot in flower colours.

For autosomal traits that arise from combinations of genes, you can able to use a Punnett square, but you need a bigger square. For example, a trait that comes from 2 genes on different chromosomes would need a 4x4 square, and the probability of passing on each copy of a gene drops to 1/16 instead of 1/4.