Okay, I removed the first spreadsheet since that was a disgrace to proper benchmarking. In addition of using just random pub samples, it seems like the -autoconfig wasn't disabling the autoexec file properly.
Now I used 2 different demos to bench: mob's benchmark found here http://www.teamfortress.tv/7598/tf2-benchmarks and my own recording from koth_king pub. These 2 demos represent different scenarios: light map with lots of players, and slightly heavier map with less players.
Spreadsheet:
https://docs.google.com/spreadsheets/d/1UQtkFIUTI0wY57tN-1kP1deZgjFdibalwAnZN3JAo4c/edit?usp=sharing
Conclusion: Comanglia cfg works as it should, boosting the fps around 23% amongst all dx modes in my case. I lost the numbers for DX80 mode, but commonly the performance between DX80 vs DX81 and DX90 vs DX95 was very similar: in the end DX81 and DX90 beat their siblings by around 1-3%.
Don't beat your siblings.
Apologies.