In a recent game, I forked a queen and rook with a knight. After seeing no immediate threats from 8. Qf4 I immediately captured with 8. ...Nxb3. Apparently this was quite a mistake (a 213 centipawn evaluation drop), but it's not clear to me why the material advantage from Nxb3 is inferior to Stockfish's recommendation of Nxc2, let alone far superior.
[White "White"]
[Black "Black"]
[fen ""]
1. e4 c5 2. Bc4 { B20 Sicilian Defense: Bowdler Attack } b6 3. a4 Bb7 4. Qf3 e6 5. Ra3? { (-0.77 → -2.15) Mistake. d3 was best. } a6 6. Rb3 Nc6 7. Nh3? { (-2.78 → -5.30) Mistake. Re3 was best. } Nd4 8. Qf4 Nxb3? { (-5.02 → -2.89) Mistake. Nxc2 was best. } 9. Ng5?? { (-2.89 → -6.87) Blunder. Bxb3 was best. } f6 10. Nxe6 dxe6 11. Bxe6?! { (-5.59 → -10.58) Inaccuracy. Bxb3 was best. } Nd4 12. Bg4 Nxc2+ 0-1
What am I missing here? Why shouldn't Nxb3 be played (or maybe, why is Nxc2 far superior)?