Nice that you mention 17...Bxd3. When I was asked to give some variations yesterday, I had another look at the line to check what I might have missed. My attention also went to 17...Bxd3 and a game with it I had in my database. It went:
18.cxd3 Nd719.Nd5, and black had to give a pawn after 19...Qxd2 (19...Qd8 must be horrible, e.g. 20.Bg5) Rxd2 since both a7 and e7 are hanging.
So 18...Nd7 can not be the solution, if there is any after 17...Bxd3 at all. I therefore had a look at the computer suggestion 18...Rc7, which fails to impress after 19.Bh6 Bh8 20.g4! and black cannot use the extra tempo. I looked at 20...Rbc8, 20...Qe5 and 20...b5, and they all lose.
Perhaps instead 18...b5 is an option. Compared to 17...b5, the bishops are exchanged and the c file is a bit more opened. Black now has a line at his disposal which is bad in the line with 17...b5:
18...b5 19.Bh6 Bh8 20.Bf8? Nh5! and now after 21.Rxh5 there is 21...Bxc3 22.bxc3 Qxc3, which would not work of course with a black bishop on c4. Black is at least ok in that position. Also after 21.Bxe7, which is an interesting option in the 17...b5 line too, black can get away with 21...Bxc3 22.bxc3 Qa3. Perhaps he is even a bit better.
Instead 20.Bg5 Bg7 21.Bh6 Bh8 is a repetition, when the white attempt to play for a win with 21.d4 does not change much after 21...b4 22.Na2 Qa4 23.Nxb4 a5 when perhaps white should try to force a draw with 24.Bxf6 axb4 (24...Bxf6!?) 25.Bxg7 bxa3 26.Rh8+ =
In my vies the most promising move is 20.Nd5, since it forces 20...Qd8. Nevertheless I have not found a way for white to get an advantage in this position, so indeed the "delayed b5" after 17...Bxd3 18.cxd3 seems to be ok for black. Perhaps white has to try 18.Qxd3 instead but of course it looks strange after 17.Be3. Suggestions are of course welcomed