why splice() performs so bad on my system?

I want to test the performance of the splice() syscall. I compare it with the traditional read/write.

/* wr.cpp 
 * it use read/write
 */
#include  <sys/types.h>
#include  <sys/stat.h>
#include  <fcntl.h>
#include  <unistd.h>

#define BUF_SIZE 4096

int main(int argc, char *argv[])
{
    char buf[BUF_SIZE];
    int in = open("1.rmvb",O_RDONLY);
    int out = open("1.cp.rmvb",O_WRONLY|O_CREAT,0766);

    ssize_t nread;
    while( (nread = read(in,buf,BUF_SIZE)) > 0 )
    {
        write(out,buf,nread);
    }

    return 0;
}

//

/* splice.cpp 
 * it use splice
 */
#ifndef _GNU_SOURCE
#define _GNU_SOU开发者_如何学GoRCE
#endif

#include  <sys/types.h>
#include  <sys/stat.h>
#include  <fcntl.h>
#include  <unistd.h>

#define BUF_SIZE 4096

int main(int argc, char *argv[])
{
    char buf[BUF_SIZE];
    int in = open("1.rmvb",O_RDONLY);
    int out = open("1.cp.rmvb",O_WRONLY|O_CREAT,0766);

    ssize_t nread;
    while( (nread = splice(in,NULL,p[1],NULL,BUF_SIZE,0)) > 0)
            splice(p[0],NULL,out,NULL,BUF_SIZE,0);


    return 0;
}

here is the result:

It seems that spilce() didn't improve the performance and didn't reduce the CPU time. Why? My kernel version is 2.6.35-28, ubuntu 10.10.

Are you sure, that one of your descriptors is actually a pipe? Because, man splice says:

... It transfers up to len bytes of data from the file descriptor fd_in to the file descriptor fd_out, where one of the descriptors must refer to a pipe.