What you can do with *paleontology*? You never will find courtship behaviours, colours of the most (oh wait!)... The methodology is sound enough to molecular biologists use this techniques and reconstitute ancestral sequences.
First, are you sure you haven't misunderstood the technique? Are you sure it doesn't involve phylogenetic bracketing right from the start, so that you need a bird _and_ a crocodile -- or, rather, the ancestral neornithean and the ancestral (crown-)crocodylian sequence -- _before_ you start comparing the *Tyrannosaurus* aa sequence to anything?
Because that way, you would _inter_polate. What you did was to _extra_polate.
Second, it has happened before that molecular biologists used unsound techniques. The literature on molecular dating is full of papers that use completely erroneous calibration points (and too few of them), and even fuller of papers that use all calibration points as maximum ages and thus arrive at much too old dates.
The rationale is basically the same as the morphological phylogenies. We can infer that the most recent common ancestor of protostomes and deuterostomes had bilateral simmetry. There are many characteristics that are inferred for the Urbilateria. Even for Cenancestor or LUCA.
Absolutely not. All conclusions about them are drawn from _inter_polation: you compare descendants and draw conclusions about their common ancestor. You do not go beyond that.