Um... If they really cared about making a quality product they would fix their existing problems, including their (apparently) poor choice of a license. Just because it is hard doesn't mean their shouldn't fix it.
It's not that it is hard, it is that there are literally thousands of contributors that have contributed code under a gplv2 license that would have to be hunted down and consulted about any license change. I'm sure if you're up for it they would be happy to have someone give it a shot so that if a re-license was needed in the future they could do it, but right now there is simply no way to legally re-license the code without spending thousands of hours getting people to sign little pieces of paper that could be used improving the kernel (zfs or no).
Right. So hard? Its a lot of work that is unplesant and nobody wants to do, I think that qualifies as "hard". I do agree its impractical, but it seems that for something as important as a function, working file system (and don't suggest that butterface fs is a production ready alternative) that somebody would put in the work to make it happen. After all the file system is everything, if you can't safely store your data, whats the point?
As an addition to the above, clearly ZFS is better than the common alternative: raid cards running some unknown firmware (that is most definitely open).