yappi overhead is proportional to the number of function calls in the codepath. So yappi could penalize some code paths more than some other code paths.

gistfile1.py

import yappi
import time
 
yappi.set_clock_type("cpu")
 
def foo(n):
   s = 0
   for i in xrange(n):
      s += n*n

def bar(s,n):
   return s + n*n

def baz(n):
   s = 0
   for i in xrange(n):
      s = bar(s,n)

def profile_foo(n):
   yappi.start(True)
   foo(n)
   yappi.stop()
   yappi.clear_stats()


def profile_baz(n):
   yappi.start(True)
   baz(n)
   yappi.stop()
   yappi.clear_stats()

def time_foo(n):
   start = time.time()
   profile_foo(n)
   print time.time() - start


def time_baz(n):
   start = time.time()
   profile_baz(n)
   print time.time() - start


def time_pure_foo(n):
   start = time.time()
   foo(n)
   print time.time() - start


def time_pure_baz(n):
   start = time.time()
   baz(n)
   print time.time() - start

time_baz(50000000)