[macbri]

You've definitely found something -- running valgrind on awk with your example and again on GNU awk (from MacPorts) shows that sure enough awk leaks memory like a sieve, while GNU awk doesn't.

It might be worth filing a Bug Report with Apple (not that you'll ever hear back but it might at least call their attention to it and get it fixed in a future release).

Consider installing GNU awk, which is not only not afflicted by this apparent bug, but also more powerful. As a quick fix, if you don't want to install GNU awk right now, you can modify your code as shown below, and even the Apple-provided awk doesn't leak memory with this. There's one major caveat -- you're not guaranteed to have your array items in any particular order, so it may not be suitable for what you want to accomplish:

# Store the first 800 words 
NR < 801 { word[NR] = $1 }

# Now access a stored array value without a memory leak
NR > 800 {
    for (i in word)
        printf("%d\t%d\t%s\n", NR, i, word[i]);
}