Probably OT (but as you mentioned SDCC) instead of using the unions of AN155 and the lengthy (and not portable) stuff:

typedef union                          // union used for writing to TL0 & TH0
    {
        struct
        {
            unsigned char hi;
            unsigned char lo
        } b;
        unsigned int w;
    }udblbyte;

...
      TR0 = 0;                         // stop Timer0
      time.b.lo = TL0;                 // read lo byte first
      time.b.hi = TH0;                 // read hi byte second
      time.w = time.w - offset;        // calculate new time 
      TL0 = time.b.lo;                 // write lo byte first
      TH0 = time.b.hi;                 // write hi byte second
      TR0 = 1;                         // start Timer0

 

it also could be done like this:

      TR0 = 0;                         // stop Timer0
      TMR0 -= offset;
      TR0 = 1;                         // start Timer0

 

instead?

Compiles to:

                            609 ;       stepper.c:7: TR0 = 0;                         // stop Timer0
   0007 C2 8C               610         clr     _TR0
                            611 ;       stepper.c:8: TMR0 -= offset;
   0009 E5 8A               612         mov     a,_TMR0
   000B C3                  613         clr     c
   000C 95*00               614         subb    a,_offset
   000E F5 8A               615         mov     _TMR0,a
   0010 E5 8C               616         mov     a,(_TMR0 >> 8)
   0012 95*01               617         subb    a,(_offset + 1)
   0014 F5 8C               618         mov     (_TMR0 >> 8),a
                            619 ;       stepper.c:9: TR0 = 1;                         // start Timer0
   0016 D2 8C               620         setb    _TR0

 

Which is close to the metal.

If you want to reproduce, SDCC compilable code is:

#include "msc1210.h"
extern int offset;
void Timer_ISR (void) __interrupt 1
{
      TR0 = 0;                         // stop Timer0
      TMR0 -= offset;
      TR0 = 1;                         // start Timer0
}