Once the components are parsed into whichever data structure is used internally (by the rendering library), wouldn't the memory and processing footprint be the same either way? I do agree, it would make the files much larger, but as pointed out in another comment, that should be a clear indication that XML simply isn't a good fit for this use case.
btw - I use SVG a fair amount myself, and have written my own path parser and builder recently. While it's not inherently too difficult, is definitely is annoying, especially when the format of the file (XML) should be able to accommodate the path components.
Semantically, if child nodes can be used for composition (e.g., a document element with child nodes that describe the document contents), then I don't see how any SVG <shape> should be different.