why are java RandomAccessFile so much slower than FileOutputStream?

Question

As long as I can understand java api, opening a RandomAccessFile using "rw" does not write ervery single byte synchronously on the underlying storage device. Unlike with "rws" or "rwd".
Why is it almost the same "speed" like the unbuffered FileOutputStream with "rw" and more than 10 times slower with "rws"/"rwd"?

the following simple code shows this, and I cannot get any reasonnable explanation to this. Any clue?

import java.io.File;
import java.io.FileOutputStream;
import java.io.OutputStream;
import java.io.RandomAccessFile;

public class StreamTest {

  public static void main(String[] args) throws Exception {

    OutputStream os;
    RandomAccessFile raf;
    int size = 10000;
    File file = new File("test.log");

    long a=System.currentTimeMillis();
    os = new FileOutputStream(file);
    for(int i=0;i<size;i++){
      os.write(("1").getBytes());
    }
    os.close();     
    long b=System.currentTimeMillis();
    System.out.println("writing direct "+(b-a));

    raf = new RandomAccessFile(file,"rws");
    for(int i=0;i<size;i++){            
      raf.write(("1").getBytes());
    }
    raf.close();

    long c=System.currentTimeMillis();
    System.out.println("random access write "+(c-b));

    raf = new RandomAccessFile(file,"rw");
    for(int i=0;i<size;i++){            
      raf.write(("1").getBytes());
    }
    raf.close();

    long d=System.currentTimeMillis();
    System.out.println("random access optimized write "+(d-c));
  }
}

Answer 1

From the docs, rws mode means:

Open for reading and writing, as with "rw", and also require that every update to the file's content or metadata be written synchronously to the underlying storage device.

It stands to reason that forcing the file's content to be written to the underlying device would be significantly slower than the other methods, which probably allow the VM/OS to cache the writes.