I was able to save at the end of ch1-1 and got a time of 3:32 with my current WIP, which is only 2 seconds slower than the SDA NewGame+ record. The NG+ also used Ditman to save about 10 seconds ahead of me in the first area alone. With that said...
Watching the two segments side-by-side, my TAS was 3 seconds real time slower than the console (close enough, 3:30 compared to 3:32). Removing load times and comparing gameplay only, the console run is about 9 seconds ahead. So, my conclusion is that the game timer does indeed include load times. It seems each load screen instance saves about 1 second on Dolphin.
Comparing my run to another New Game run in 3:58 for ch1-1, I am about 26 seconds faster in real time and game time. Console also uses a trick that, during loading cutscenes, you can pause the game with START and the game timer stops but still loads the cutscene. This saves the console about 5 seconds game time. Comparing both of these runs, it seems the loading time range is somewhere around 5-8 seconds for ch1-1. Here is that New Game run:
http://www.youtube.com/watch?v=1SXMoIeTlTA
So why am I so focused on the game timer when TASvideos uses frame counts? Well some TAS runs go for game time (several Metroid games do). Also, if each load screen saves 20-30 frames regardless, this will result in a run several minutes faster than it should be. Anyone who has played RE4 knows that there is a metric shit-ton of load screens.
Zanoab is right in his assumption that an ISO would load faster than a disc. Sadly this is the obvious truth. For games like the other Wii submission MegaMan10, the whole game can probably fit into the RAM of the system, so load times don't exist.