Okay, now about the movie itself.
While this run is
much better than the previous submission, it still looks like you mostly played it regularly with a few retries, instead of optimizing every action to gain you as much time as possible.
The most obvious thing is driving in the middle or in the outer side of a curve instead of trying to cut the overall distance you travel.
For every turn, you should aim to move on the innermost side as close to the grass/ground as possible, until you start losing speed (green line). If the turn is too sharp and you can't keep near the innermost side all the time, at least touch it while turning as much as you can (orange line). If you just move in the middle or on the outermost side, you simply travel more distance, losing time (red line). I tried to do this all the time in my test movies, and I clearly see you're not doing it here. Maybe it doesn't matter due to limitations of this game? I might check that. Also when you brake to allow your car to turn more sharply, you're losing speed, so for every turn that you can't clear as orange, you have to find balance between braking and releasing the X button.
Then, judging by the rerecord count, you haven't optimized every gear shift to gain you as much speed as possible.
Also in some places it looked like you could still bump the main player car with some nitro, but you just kept standing still or driving slowly, wasting nitro on nothing or not using it at all. You mention that sometimes you have to waste the fuel for the race to end, and you don't have time to bump your opponent, but it feels like this hasn't been tested too hard.
Overall I like how this movie turned out, and improvement of several minutes tells quite clearly that the strats are way better this time around, but the problems I mentioned still stick out as sub-optimal.
The reason I'm suggesting bizhawk is that you have a tool that's incredibly useful for tight optimization, as well as for movies with several players acting independently - TAStudio. It also allows you to edit your past input as much as needed, while not erasing the future events, like usual rerecording does. So if you stick to just old-school "savestates only" approach, and you aren't too good at it, you will always end up with sub-optimal movies that can be beaten easily with just better tools.