In defense of "the netbeans guys", they didn't actually write that code (although that is what the performance issue resolved to).
An instance of FilterOutputStream was used in between somewhere and whoever implemented it missed the comment in the JDK docs that says "The write method of FilterOutputStream calls the write method of one argument on each byte to output. ... Subclasses of FilterOutputStream should provide a more efficient implementation of this method. " The code that copied files for deployment was just writing to an output stream retrieved from FileObject.getOutputStream() on the assumption that it was correctly configured.
Still disappointing, but maybe it's easier to see how it happened.
[Message sent by forum member 'peterwx' (peterwx)]
http://forums.java.net/jive/thread.jspa?messageID=284523