The existing implementation of Min/Max calls InOrderTreeWalk,
which perform an allocation of a stack for traversing the tree.
But when you are finding the smallest or largest element in a
tree there is no need to perform this allocation. This is an
optimised version that directly finds the smallest or largest
element.

This is an important use case for using a sorted set as a priority
queue. In microbenchmarking, with the following code I
found a 30-40% speed up:

        for(int j=0; j < 10; j++) {
            var s = new SortedSet<object>();
            for(int n = 0; n < 100000; n++) {
                s.Add(n);
            }

            int i = 0;
            while(s.Min != null) 
            {
                var o = s.Min;
                s.Remove(o);
                i++;
                if(i%2 == 0) {
                    s.Add((int)o * 2);
                }
            }
        }