InkML <trace> element; why require a BNF parser in addition to an XML parser?

The <trace> element in InkML seems to abandon XML in favor of a custom 
BNF parser, I assume to try and save space.  I was wondering, what is 
the rationale behind this decision?  Interoperability, ease of 
implementation, and extensibility seem to dwarf concerns about space 
efficiency in almost all widely-used protocols on the internet.  InkML 
as-is is like two different languages, and requires two different 
parsers.  Why not bite the bullet on space concerns and use XML in the 
<trace> element, and ditch things like 'a' to 'y' being interpreted as 
decimals?