XSLT Version 2.1 Draft published. As mentioned in the published draft the main goal is making streaming of transformation easier for implementors.
Streaming makes life easier whenever you have to process huge input streams of information to reduce memory footprint and deliver already processed chunks to the next consumer before the complete stream is successful processed.
On the other hand XSLT 2.0 which is a recommendation since 3 years is not widely adopted from implementors of XSLT engines -- Saxon, AltovaXML and few commercial ones-- so far.
Why? The main reason is complexity compared to the corresponding benefit. The implementation of a XSLT engine passes the 2.0 conformance tests is expensive and complex compared to the benefit for most users / use-cases. As in any commercial product each new feature / requirement should be qualified against the value and cost / effort it brings into the standard.
The interesting question would be -- how to measure benefits of a feature? How to interview the "users" of a standard? There is no trivial answer and that might be the main reason why standards tend to get over complex.