This is not a change we'll want to make standard, though. It would be
nice to have a way to tell the template parser to put out the page in "chunks", but in the normal case we don't want to do this.
As Tom mentioned, ns_write is *expensive*. In a test I did a very long time ago (a couple of years ago) I was able to triple the throughput of pages out of my old P200 box by assigning data gotten from the db to a "page" string variable later returned by ns_return,
vs. using ns_write to put the data out a row at a time.