The least memory intensive solution would be to read over the source
document with an XmlTextReader and switch to a different output ( presumably
a file ) every so often.

<pseudoCode>

XmlTextReader reader = new XmlTextReader( ... );
XmlWriter writer = new XmlTextWriter( GetNextFilename() )
int pageSize = 100; // chunk every 100 rows
int currentRow = 0;

// Init the writer
writer.WriteStartElement( "myRoot" ); // because you can't write fragments
with the XmlTextWriter

reader.Read() // skip over the top level element

while( reader.Read() )
{
   if( 0 == ((++currentRow)%pageSize) )
   {
       writer.WriteEndElement();
       writer = new XmlTextWriter( GetNextFileName() );
   }
   writer.WriteNode( reader, false );
}

writer.WriteEndElement();
</pseudoCode>

Using the Reader / Writer combination you never have to read the entire XML
document into memory, which was probably the concern in the first place.

Alternatively, you might shred the XML document into SQL Server tables using
SQLXML bulkload and then use queries to retrieve subsets of the data.

Is there any way you can just reduce the size of the XML document in the
first place or are you just receiving those huge documents.