Not really a great "delay" implementation. If someone were to use such a crude looping system to delay a program (and I doubt any semi-modern operating system does this) some NOP instructions would be much simpler.

Real delay operations context switch to another process or thread for a minimum of the desired amount of time. A better implementation for single-tasking systems would be something like the following pseudocode:

    ; we assume the realtime clock is already setup to trigger timer interrupts at a regular interval (short enough that our delay function has the desired precision)
    delay:
        ; Assume time to delay (in timestamp ticks, converted from milliseconds elsewhere) is in ebx
        RDTSC eax ; Read timestamp into ebx (not the real way to use RDSTC!)
        ADD ebx, eax ; ebx now contains the timestamp coutner value of when the delay should end
    .delay_loop:
        CMP eax, ebx ; compare the current timestamp to the end time
        JG .delay_end ; if the current time is greater than the end time, we're done
        ; Otherwise we wait...
        HLT ; Put the processor in lower power "sleep" state
        ; When the timer interrupt fires (at regular intervals), the processor is woken up
        RDTSC eax ; Read new timestamp counter value (not the real way to use RDSTC!)
        jmp .delay_loop ; Compare again and keep waiting 
    .delay_end:
        RET ; return from delay function