I improve Finding Nemo by implementing a glitch already known in the RTA record where you can re-use a pearl that is ordinarily eaten by a clam by pausing and unpausing with the correct timing. This saves ~20 seconds overall.
This game has some level of RNG or some unknown global state that effects collisions. For example the flag in the first fish tank stage moves to the left, which is slower, instead of the right as in the original. I was unable to get it to move in the correct direction. This also effects the level where you have to stop the spinning blades with a pearl, and it desynced pretty badly. So a few seconds were lost in these two levels. Some RNG actually worked in my favor though in the first treasure chest stage which is now about 1 second faster. I additionally saved a couple seconds in the next to last stage by swimming normally against the current instead of dashing. For the current moving at this particular speed it is actually much faster to not dash.
You might notice that this run appears to be only 7 seconds faster than the original, but this is due to the original not using the BIOS and the fact that VBA doesn't count all lag frames individually. If you look at the encode time for the original you will see it runs for ~17:55 not the stated time of 17:47. Adding in BIOS this would be ~17:59. The bulk of this difference comes from lag in the ring chase stages. Even more lag is added here in GBAHawk, eating up some of the savings.
This run is console verified, as in the temp encode.