He doesn't say why in the short one (I haven't watched the long one). Late mergers are definitely bad in some situations - for example, when they are merging into a slow-moving turnoff lane, blocking a whole lane of ongoing traffic.
I agree though that it doesn't really work in all situations. Tom uses the example of two lanes merging into one for roadworks. Maybe some situations where it doesn't work are an example of poor road design by traffic engineers?