Big numbers, strange errors

#1
I have a @FACT plugin (factorial), essentially:
Code:
{
   ...
   Evaluate(psz);
   INT x = _wtoi(psz);
   while ( x > 1 )
   {
     Sprintf(psz+lstrlen(psz), L"*%d", (x -= 1));
     Evaluate(psz);
   }
   return 0;
}
It works fine up to %@FACT[3248] when he result (and Evaluate()'s argument) are just under 10000 characters (@eval's output limit for integers). So I limit @FACT to numbers <= 3248. That works fine.

I'm working on another plugin, @BINOM[n,k] (binomial coefficient, n!/(k!(n-k)!)). Right now,
Code:
INT WINAPI f_BINOM ( WCHAR *psz )
{
   DWORD n, k;
   if ( Sscanf(psz, L"%u,%u", &n, &k) != 2 || n > 3248 || k > n )
     return 0;
   Sprintf(psz, L"%%@FACT[%d] / %%@FACT[%d] / %%@FACT[%d]", n, k, n-k);
   ExpandVariables(psz, 0);
   Evaluate(psz);
   return 0;
}
If I try @BINOM[3200,1600] (about 1000 digits), it works. The expanded line passed to Evaluate() is about 18700 characters long and its first (and longest) token is about 9830 characters long. If I try it a second time, any of a variety of things may happen. One result (which I've never seen before) is:
Code:
v:\> echo %@binom[3200,1600]
MAPM Error: 'm_apm_to_fixpt_string', Out of memory
After that, a third try produces nothing
Code:
v:\> echo %@binom[3200,1600]
v:\>
And ~30 seconds later, TCC disappears.

I don't think I'm running into TCC's size limits. But I'd like to limit it to avoid errors.
Does anyone (Rex?) have an idea of what difficulty it's running into?