Improving on Fischer, Geller, and Kasparov?

The key position in Fischer - Geller, Skopje 1967 is shown in the following diagram (see that previous post for the PGN and a link to Chessgames.com). Fischer played 20.a3, where Fischer, Geller, and Kasparov all gave the move a '?', suggesting 20.Qf4! instead.

Skopje 1967
Geller, Efim

Fischer, Robert
After 19...Nxe4

My engine finds 20.Qf4 better for White at depth 13, giving 20...cxb2 21.Rh5 Nf6 {FIS: 21...Nc3+} 22.Rh6 Rxf7 {GEL: 22...d5} 23.Bxf7 Be4 24.Bb3 d5 25.Bxf6 Bxc2+ {KAS: 25...gxf6}, as best. I've indicated where the machine deviates from the published analysis of the three GMs, who each in turn improved on the other's previous analysis.

Now the machine comes up with 26.Bxc2 Qxf4 27.Rxh7+ Kg8 28.Rxg7+ Kh8 29.Rh7+ Kg8 30.Bxe7 Rc8 31.Rh3 Rxc2 32.Kxc2 Qc7+ 33.Kb1 Qxe7 34.Rxd5 Qe4+ 35.Rdd3 Qe1+ 36.Kxb2 Qb4+ 37.Kc2 Qc4+ as best. I've left out side variations that, although significant, don't change the course of the game dramatically.

In the final position, Black is behind in material with 2R+3P vs. Q+P, but the White King has difficulty finding shelter from the Queen's checks without dropping a Pawn. I abandoned the analysis here because I was far from the diagrammed position ('long analysis, wrong analysis') and finding a technical win for White would take more time than I had available. While Black is behind materially, it's worth remembering that 2R+P vs. Q is often a theoretical draw.