That's actually the point of the algorithm. It's actually enlarging or shrinking regions on the shadow map based on how important they are to the final scene-- in the case you mention, the RTW warp maps would 'steal' some resolution from the surrounding/unimportant areas (say, stuff cast on scene backfaces or not in the view frustum outright) and use that for areas that are actually visible. With some tweaking, you could probably get uniform sampling density across the image!
Re: amortization-- True, this is an advantage of cascaded shadow maps. On the other hand, you don't need to do extra processing to determine which shadow map slice your objects will occupy or doing edge case handling when it straddles boundaries. Profile for your use case?
EDIT: There are some trees towards the end of the vid-- I think they hold up pretty well
clb: At the end of 2012, the positions of jupiter, saturn, mercury, and deimos are aligned so as to cause a denormalized flush-to-zero bug when computing earth's gravitational force, slinging it to the sun.