tee has a simple main loop:

  while not EOF: 
    (synchrounously) read up to BUFSIZE bytes from stdin into buf 
    for each output file: 
      (synchrounously) write buf to file

(source code here: http://git.savannah.gnu.org/cgit/coreutils.git/tree/src/tee....)

All of the reads and writes are done through synchronous file apis. I don't see where you are getting one-thread-per-file (or any other kind of parallelism) there.