Following my working code for the AT89C4051.
It use T0 in mode 3 because T1 was used for the UART.
Thus TL0 and TH0 are cascaded to 16 bit inside its interupt handlers.
For stable results the whole measuring was done over several periods. A value of about 200 * 256 CPU-cycle seems sufficient.

#pragma src(measure.a51)
#include <stdio.h>
#include <math.h>
#include <reg4051.h>

#define	XTAL	11.0592e6
#define BAUD	4800

typedef unsigned char u8;
typedef unsigned short u16;

u8	period_h;
u8	period_l;
u8	pulse_h;
u8	pulse_l;
bit	conversion_done;
u8 tl0_h;
u8 th0_h;


void int_tl0( void ) interrupt INT_T0	// count pulse
{
  tl0_h++;
}


void int_th0( void ) interrupt INT_T1	// count period
{
  if( ++th0_h == 200 ){
    IE0 = 0;
    EX0 = 1;				// enable next falling edge
  }
}


void int_ex0( void ) interrupt INT_EX0
{
  if( PX0 ){
#pragma asm
	xch	a, pulse_l		; save A
	clr	a
	xch	a, th0			; read and clear
	mov	period_l, a
	clr	a
	xch	a, tl0			; read and clear
	xch	a, pulse_l
#pragma endasm
    IE0 = 1;				// force calling again
    IP = PT0_ | PT1_;			// handle timers first
  }else{
    period_h = th0_h;
    th0_h = 0;
    pulse_h = tl0_h;
    tl0_h = 0;
    conversion_done = 1;
    EX0 = 0;				// disable
    IP = PX0_;				// restore priority
  }
}


void main (void)
{
  u16 period, pulse;
  PCON |= SMOD_;
  SCON = TI_ | REN_ | SM1_;
  TMOD = T0_M0_ | T0_M1_ | T0_GA_ | T1_M1_;
  TH1 = (u8) -(0.5 + XTAL / 12.0 / 16 / BAUD);
  TL1 = -1;
  TCON = TR0_ | TR1_ | IT0_;
  IP = PX0_;
  IE = ET0_ | ET1_ | EA_;

  printf( "Hallo Petern" );
  for(;;){ 						// main loop
    conversion_done = 0;
    while( conversion_done == 0 );
    period = (u16)period_h << 8 | period_l;
    pulse = (u16)pulse_h << 8 | pulse_l;
    printf("%8u", period );
    printf("%8u", pulse );
    printf("%8.2fn", 212.766 * ((float)pulse / period - 0.32) );
  }
}

Peter