updated FAQ for performance numbers.  I guess I should remove that point anyways.

--HG--
branch : trunk
diff --git a/docs/faq.rst b/docs/faq.rst
index dc75f98..3a04c20 100644
--- a/docs/faq.rst
+++ b/docs/faq.rst
@@ -18,8 +18,8 @@
 We really hate benchmarks especially since they don't reflect much.  The
 performance of a template depends on many factors and you would have to
 benchmark different engines in different situations.  The benchmarks from the
-testsuite show that Jinja2 has a similar performance to `Mako`_ and is more
-than 20 times faster than Django's template engine or Genshi.  These numbers
+testsuite show that Jinja2 has a similar performance to `Mako`_ and is between
+10 and 20 times faster than Django's template engine or Genshi.  These numbers
 should be taken with tons of salt as the benchmarks that took these numbers
 only test a few performance related situations such as looping.  They are
 not a good indicator for the templates used in the average application.
diff --git a/jinja2/environment.py b/jinja2/environment.py
index 5ec8cb5..dc8bc25 100644
--- a/jinja2/environment.py
+++ b/jinja2/environment.py
@@ -654,11 +654,10 @@
         if size <= 1:
             raise ValueError('buffer size too small')
 
-        def generator():
+        def generator(next):
             buf = []
             c_size = 0
             push = buf.append
-            next = self._gen.next
 
             while 1:
                 try:
@@ -675,7 +674,7 @@
                 c_size = 0
 
         self.buffered = True
-        self._next = generator().next
+        self._next = generator(self._gen.next).next
 
     def __iter__(self):
         return self