My experience with large XML based systems, is that the power rerquired at the Web server end is underspeced and the controlling body then reverts to 'offline' working, which makes a bit of a mockery of the ideal that when the last byte of the xml stream is sent, you immediatly know it is valid and that it matches the schema.
EMAsys couldn't cope with the data sent 'live',so they introduced version 2 which has offline working built in.