@TomV, @AndriyM I looked more closely on the execution plans when using flwor (loop) and a path expression. On the surface the plans are identical but there is a difference in one compute scalar (don't know why it is there) that makes the loop version substantially slower.
@AndriyM Yep, percentages are even more useless than usual when it comes to XML queries. BTW, I attended SQL Saturday a couple of days ago and a well renowned speaker used the percentages to show difference in performance between compatibility levels.
@MikaelEriksson Interesting. Does that speaker have a blog, by any chance? Are their blog articles known, perhaps, to often come up in the top results in search engines?
