Oh_Deer Explanation

Between darb and myself, we figured out lots of small strategy improvements, without which the level of time improvement that should follow, would not have been possible. Just before darb's entrance, I had been working on a deep empirical analysis on lag, level by level. I was able to improve the existing level strategies in that regard, which amounted to a possible time save of almost 2 seconds over the course of the run. Darb discovered that certain map movements (after boss fights) as well as entering the very first level could be buffered. It was also darb who suggested and pioneered a small powerup route change in the submarine and the whale level, which saved close to a second. I then discovered a new swimming technique with bunny ears which not only speeds up the movement in certain rooms of the level, but also provides a consistent and faster setup for darb's new powerup route. Darb also suggested an alternative powerup route in Mario 3 and 4, but, initially, we could not make it work right, so we ended up not saving time with it. In comes the 2020 Glitchless tournament for which Controllerhead reappears from his 6 year hiatus and suggests the same route change. I went into the speedrun laboratory once more and figured out the right movement to make the strat save time on average, up to 0.4 seconds, if done perfectly.
Most importantly, the big game changer and motor for the rapid improvement of every active runners' times in 2020 was the development of the practice rom by lightbulbsun. I'm confident in saying that almost 10 seconds of the time saves that would occur since the 26:27 are thanks to the drastically improved practice conditions that came with its introduction.
Lightbulbsun even provided a dedicated practice rom just for the Tatanga fight - along with a detailed explanation of what he had found out about the fight's mechanics through code analysis. With that knowledge and practice tool, we were able to finally understand the specifics of the fight and get more consistent at the so-called 1-cycle, which had been possible before, but was a rather rare outcome before we actually understood what we had to do. A 1-cycle saves 1.5 seconds over the regular 2 cycle. Side note: Even a 0-cycle is possible with almost inhuman inputs - and by 2021, it's been pulled off by at least 4 people, but is still considered too inconsistent for actual runs.
Later on, Adam would also suggest small level strat improvements here and there and generally push execution to the limit. Lightbulbsun would continue to look into the game's code to help with strats and slashinfty, the maker of the current TAS, the SML2 randomizer and former any% wr holder would identify some new lag strats.
Bottom line: 2020 was a great community effort in optimization and creative thinking. From ~2014 to 2019, there had been a few things optimized here and there. Compared to that, 2020 was a relative explosion of ideas and discoveries.