There are some really good piano samples available for free--why put in the work for the physical modeling approach?
In (very) short: the resonance (notes interacting with one another), combined with the analog nature of varying velocity (how hard keys are pressed), combined with pedaling modifying both of the above.
The resulting combinatorial explosion means that the number of samples you need to capture, in order to have a high-fidelity reproduction of a physical piano, is enormous.
Putting aside the practicality of capturing all of this, you're still looking at tens to hundreds of gigabytes of raw samples per piano.
(Great question, by the way.)
Pianoteq has been around for 17 years -- as long as patents last! -- and yet open source still seems to have nothing that comes close.