I would not say it was fully sound, but having three pawns for that piece, with Black's pieces largely undeveloped, seems very reasonable for compensation.
A computer might give a poor evaluation, but of course we cannot expect to play like a computer. Some winning lines, for example, are very computerish and we may not always be confident along those lines or may not even see them at all.
In game one, the text book example, I think Black could have stayed in the game if he had played 16...Ke7 instead of castling. This doesn't make White's sacrifice wrong. He gets excellent practical chances for the piece. But putting the black king as far from the queenside as possible seems wrong.
In your game, 12...Kf8 would be hard for me to find too. But I can see the logic. White's compensation is those three queenside pawns, and the more pieces that come off the board, the stronger those pawns get. So Kf8 avoids a bunch of exchanges and Black can hope to make his extra piece count
My engine doesn't agree that 10. Bxb5 is a brilliant move in the first game. Bxb5 isn't losing, but it isn't winning, either. Bd3 and Bxb5 are equal. a3 is pretty darn close.
Personally, I don't like the idea of giving my opponent half open files to start attacking my castled position. As black, I'd gladly trade those pawns for your bishop. What you have going for you in your game is your opponent is behind in development. He's either going to leave his king in the middle or castle kingside. Since you've already castled queenside, I'd take that as an invitation to start a pawn storm attack with your g and h pawns.
So I guess what I'm saying is the first game is an example of a move that maintains equality. In your game, you could have had an advantage, and your move also pushed the position toward equality. So in that sense, you followed the pattern.
"the first game is an example of a move that maintains equality"
There are a lot of positions that are "equal," but quite a bit easier to play for one side or the other. In game 1, I think most masters would rather play White
I think the difference in evaluation is simply because you have better plans in game two than an imbalance of bishop for three pawns. 10. Bxb5 is not bad, but it's also not winning (0.00 in evaluation).
In the second game, your queen and bishop are well-placed in the center instead of blocking your king-side pawns, black is underdeveloped especially its queenside, and your queen rook battalion is eyeing the king, so they have to castle kingside. All these factors make a king-side pawn storm 10. g4 very strong in this position.
Thank Toy for You answers. After doing some more thinking and toying with the engine I too think that the sacrifice was sound, but simply unnecessary. There where stronger continuations, the sacrifice threw away the adventage.
@LaskerNephew 16... o-o is commented in the book as the losing mistake. The King was needed to help stoping the pawns, with the text move he is too far away.
As usual in this section, I would like to ask for same help in getting a most human uderstanding of what the engine is traing to tell me. Thread's title nails the problem. Same piece is sacrificed on same square, on the same move, in the same opening, but seemingly irelevant details of each position result in drastically different engine's review.
In addidion, the evaluation of the sacrifices changed drastically depanding on depth,
I am currently reading "Sacrifices in the sicilian" by C.K Levis. He gives a following game:
So i was extremally proud of myself, when I spotted the now familiar pattern in one of my own games:
What do you think about my sacrifice? Was it sound? I suspect, that the key might be in unnatural computer continuation ...12. Kf8 avoidnig exchanges. The book states that the right strategy for white is to simplify, for black, to keep as many pieces on the board as possible.
However, the engine might also think that 10. Bxb5 just gives away the advantage, the evaluation aferward is not terrible.