performance regression in DataFrame.sum?

[cpcloud]

In [29]: df = DataFrame(randint(2, size=(8e6, 10)))

In [30]: time df.sum()
CPU times: user 1.21 s, sys: 0.03 s, total: 1.24 s
Wall time: 1.23 s
Out[30]:
0    3999325
1    3998264
2    4000047
3    3997902
4    4001078
5    4001965
6    4001701
7    4000482
8    3997581
9    4000691
dtype: int64

In [31]: time df.mean()
CPU times: user 0.38 s, sys: 0.00 s, total: 0.38 s
Wall time: 0.39 s
Out[31]:
0    0.5
1    0.5
2    0.5
3    0.5
4    0.5
5    0.5
6    0.5
7    0.5
8    0.5
9    0.5
dtype: float64

Should sum be 3.15x slower than mean?

[cpcloud]

INSTALLED VERSIONS
------------------
Python: 2.7.5.final.0
OS: Linux 3.9.9-1-ARCH #1 SMP PREEMPT Wed Jul 3 22:45:16 CEST 2013 x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8

Cython: 0.19.1
Numpy: 1.7.1
Scipy: 0.12.0
statsmodels: 0.5.0.dev-db14222
    patsy: 0.1.0
scikits.timeseries: 0.91.3
dateutil: 2.1
pytz: 2013b
bottleneck: 0.6.0
PyTables: 3.0.0
    numexpr: 2.1
matplotlib: 1.2.1
openpyxl: 1.6.2
xlrd: 0.9.2
xlwt: 0.7.5
sqlalchemy: 0.8.1
lxml: 3.2.1
bs4: 4.2.1
html5lib: 1.0b1

[cpcloud]

issue is slightly worse before #3731 so that's not the cause

[cpcloud]

Only happens for int64 dtypes so far...perf is on point for float64

[cpcloud]

new vbench for this

-------------------------------------------------------------------------------
Test name                                    | head[ms] | base[ms] |  ratio   |
-------------------------------------------------------------------------------
stat_ops_frame_sum_int_axis_1                |   0.2921 |   5.4233 |   0.0539 |
stat_ops_frame_sum_int_axis_0                |   0.3717 |   5.5280 |   0.0672 |