CC2541: Timer3 overflow irq takes longer time then expected

Part Number: CC2541

I am trying to make sense of the cc2541 clock and timer3,but for some reason I get a longer time period then expected.

My board is running with 32Mhz external oscillator and I configures the clock as follow:

CLKCONCMD = (CLKCONCMD & ~(CLKCON_OSC | CLKCON_CLKSPD)) | CLKCON_CLKSPD_32M;

And Timer3:

T3CTL = T3CTL_DIV_1 | T3CTL_MODE_FREERUN | T3CTL_OVFIM | T3CTL_CLR | T3CTL_START;

in the Time3 isr function, I toggle an output and monitor it with a scope. The scope shows a square signal at about 32khz, or a period of 32uS.

If I correctly understand the configuration, Timer3 should be incremented 32M/sec and the isr should be invoked 32M/256 /sec. resulting a period of about 8uS, which is about 4 times shorter from what I actually get.

What am I missing ?

Here is my isr

#pragma vector = T3_VECTOR
__interrupt void t3_isr(void)
{
    static uint8 on=0;
    // Clears the module interrupt flag.
        T3OVFIF = 0;
        P1_0 = on; // toggle the output
       on=!on;
   // Clears the CPU interrupt flag.
      T3IF = 0;
}