but what we can do is inline a quick hashtable lookup in the cache itself to avoid the context switch when looking up the target. This brings us down to about a 20% overhead. Now we're in the right ballpark of acceptable performance.

Now, there's one more performance thing we can do: