InterlockedIncrement vs InterlockedIncrementAcquire / Release (Redux) - c++

Question

This is a followup to this question:
I guess I don't understand the Interlocked Acquire / Release APIs. I put together the small program below. As I understand it, g_val_1, g_val_2 and g_val_3 should always be updated in the same order and should end up as all the same value. But they do not (for more than one thread).
What am I missing? Thanks.
#include "windows.h"
#include "stdio.h"
#define _THREADS_ 100
#define _TICKS_ 1000
int volatile g_threads = 0;
DWORD volatile g_val_1 = 0;
DWORD volatile g_val_2 = 0;
DWORD volatile g_val_3 = 0;
BOOL g_running = TRUE;
DWORD TestThread(PVOID ignore)
{
while (g_running)
{
InterlockedIncrementAcquire(&g_val_1);
g_val_2++;
InterlockedIncrementRelease(&g_val_3);
}
InterlockedDecrement(&g_threads);
return(0);
}
int __cdecl main(int argc, char* argv[])
{
int th, duration;
int success;
DWORD ticks;
duration = _TICKS_;
g_threads = _THREADS_;
printf("Max=%d Threads=%d Entries=%d\n", duration, g_threads);
printf("Creating Threads\n");
th = g_threads;
while (th-- > 0)
{
CreateThread(NULL,
0,
TestThread,
NULL,
NORMAL_PRIORITY_CLASS,
NULL);
}
printf("Starting Threads\n");
ticks = GetTickCount();
while ((GetTickCount() - ticks) < duration);
g_running = FALSE;
while (g_threads > 0);
ticks = GetTickCount() - ticks;
success = ((g_val_1 == g_val_2) && (g_val_1 == g_val_2));
printf("Duration=%d g_val_1=%d g_val_2=%d g_val_3=%d OK=%d\n", ticks, g_val_1, g_val_2, g_val_3, success);
}