There are some improvements, but they're mostly bugfixes. The one performance improvement that we did add (undocumented, unfortunately) was with the use of convert-by-microflows for attributes. These are no longer run in a separate transaction, which will improve the speed for very large xml imports.