dataset.readxml performance problem Service Pack 3 beta

I have a 100k xml file and I am attempting to read it into a dataset
using dataset.readxml

My xml file DOES have several nested nodes. However, none of the nodes
have more than 10 attributes. I am experiencing 10-12 seconds time to
load the xml document into the dataset.

Loading the schema first helps--but minimally (approx. 1-2 seconds)

We were previously using SP2, however, there is a known bug with SP2
where (under certain circumstances), only every other node is read
into the dataset.

Are there plans to improve the performance of the dataset.readxml
call.

Does anyone have thoughts regarding improving this performance issue.

Thanks in advance