openxml vs nodes()

Hello guys, I would like to open a thread here at sql-server-performance about the advantages and performance impacts of 2 different ways to bring xml data in a relational format.
Until now I used openxml and beside the fact that it is write-intensive and felt not really "natural" to me, it just worked fine.
Now I wanted to switch to the new xmldata.nodes() method, but I read many posts about a really bad performance of this method.
What do you think about? What are your experiences?
Abstract Example:
sp_xmlpreparedocument @doc output, @xmldata
... OPENXML ()
sp_xmlremovedocument
VERSUS
...
INNER JOIN (select POS.attr.value('@ID', 'char(32)') [UID],
POS.attr.value('@AddressType', 'smallint') [ATyp],
POS.attr.value('@Language', 'char(2)') [LangID]
from @XMLData.nodes('/Shop') POS(attr)) XD

Cheers
P.S.: Sorry for my stressed post, If you do not get the point, please let me know

I have noticed that inserting from XML is faster when using openxml and slower when using nodes(). Here is a script you can run to see the difference. The difference in performance is even more pronounced when the table is wider. Also, the execution plan for the openxml case is far simpler than for the nodes() case. I've read about supposed memory issues when using sp_xml_preparedocument (openxml), but I have not been able to force any bad memory issues (by running several scripts similar to this in parallel). I am on SQL 2005. I would love to know the results of this script on SQL 2008. Or any comments anyone has in general.
---------------------------------------------
-- Set the number of rows
declare @N int;
set @N=10;

-- Set the number of test repetitions
declare @Reps int;
set @Reps = 100;

-- create the test table
declare @TempTable table (
a int,
b uniqueidentifier,
c nvarchar(100),
d datetime,
e bit
);

-- Create some data
while (@N>0) begin
insert into @TempTable (a, b, c, d, e) values (rand(1000), newid(), newid(), getdate(), 0);
set @N = @N - 1;
end

declare @Xml xml;
set @Xml = (select * from @TempTable for xml raw('ITEM'), type, elements, root('INSERTFROMXML'), binary base64);

declare @StartTime datetime;
declare @Counter int;
declare @Elapsed1 int;
declare @Elapsed2 int;

-- Test Method 1:

delete from @TempTable;
set @Counter = 0;
set @StartTime = getdate();

while (@Counter < @Reps) begin
-- The slow part:
insert into @TempTable
select
T.c.value('(a)[1]','int'),
T.c.value('(b)[1]','uniqueidentifier'),
T.c.value('(c)[1]','nvarchar(100)'),
T.c.value('(d)[1]','datetime'),
T.c.value('(e)[1]','bit')
from @Xml.nodes('/INSERTFROMXML/ITEM') T(c)

set @Counter = @Counter + 1;
end

set @Elapsed1 = datediff(ms, @StartTime, getdate());

-- Test Method 2:

delete from @TempTable;
set @Counter = 0;
set @StartTime = getdate();

while (@Counter < @Reps) begin
-- Not so slow any more:
declare @idoc int;
exec sp_xml_preparedocument @idoc output, @Xml;
insert into @TempTable
select *
from openxml(@idoc, '/INSERTFROMXML/ITEM', 2)
with (
a int,
b uniqueidentifier,
c nvarchar(100),
d datetime,
e bit
);
exec sp_xml_removedocument @idoc;

set @Counter = @Counter + 1;
end

set @Elapsed2 = datediff(ms, @StartTime, getdate());

-- Results:

select 'Method 2 speedup: ' + convert(varchar, round(convert(float, @Elapsed1) / @Elapsed2, 1)) + 'x';

Hi,ran your script on mt SQL 2008 Dev box.@Reps = 100 gives 1.1@Reps = 100000 gives 4.1xSo thanks, Microsoft, for introducing another easy-to-use programming method that doesn't bloddy scale.I see more refactoring in my future, what fun.