The programmer in me thinks that it would be nicer to use guids for ids rather than integers. That would avoid conflicting id references between files -- like when merging two files together programmatically, inserting elements like key elements, parsing the document into a dictionary, or just searching the document. With that, I think it would make sense for element to reference folder id and image name, since ids would always be unique, but names would only be unique within a folder context.
The programmer in me also thinks that rather than storing x,y,sx,sy,r, store the first six elements of a 3x3 2d transformation matrix -- where the last three elements are assumed to be 0,0,1. Though, the tradeoff would be that it would make it less human readable, but I think its a worthwhile tradeoff, and support for skews would be a pleasant plus.
Lastly, I kinda feel like the should be a reference -- like images -- rather than a hard coded value per frame. But, that more of a feeling than having a real rational objective...
Those are my initial thoughts anyway, and these are just suggestions that cater to my programmer side. My artistic side thinks the format is structurally sufficient.