It is not clear to me how 'the standard error of the difference' was calculated
in Fig. 2 in Shell et al.
In the file FSNS_std_error_of_difference.eps, the global latitude-weighted 
area mean of FSNS ('Net solar flux at surface') for 87 (time-mean) Jan is shown.
The fig in upper-left (UL) shows the whole values of FSNS for coccoliths and control
cases. Three experiments were done for each of coccoliths and control cases
using different starting conditions.
Each experiment were done for 30 years, but the results from the first model
year were excluded, so that there are 29x3 Jan FSNS for each of coccoliths and 
control cases. In the UL fig, the Jan FSNS are arranged in the order that three
experiments were conducted in real time, and within each experiment the Jan FSNS
are arranged in the order that the model results became available.
This arrangement of the Jan FSNS was done and is consistent in both cases.
Note that the years shown on the time (x) axis are just for convenience
and don't mean the real years. The time averaged (already space_averaged) values
of the whole FSNS are independent of how the time sequences of FSNS are
created. However, the sequence of the difference of the FSNS from two cases 
depends on how the FSNS in each case are arranged in 'time' and the standard
error (deviation?) of the difference changes accordingly.
In the upper-right (UR) figure, the Jan FSNS were sorted in ascending order
in both cases. It shows FSNS in coccoliths is consistently smaller than that
in control case except last 3 Jans. However, the time averages of the whole FSNS 
are the same as in the UL fig.
In the lower-left (LL) figure, the difference of the Jan FSNS calculated from
the time sequences of FSNS shown in the UL figure.
The time mean is -0.1984 and the standard deviation is 0.822.
On the other hand, the difference of the Jan FSNS calculated from
the time sequences of FSNS (ie sorted) shown in the UR figure is shown in
the he lower-right (LR) figure.
The time mean is -0.1984 the same as in the LL figure. The standard deviation,
however, is 0.0895, one tenth of the 'unsorted' case.
I don't think the paring of Jan values in either case is justified for
calculating the 'standard error of the difference' and I'm not sure what the
correct way to calculate the 'standard error of the difference' or even
what it is.

The figs for the modified (corrected) albedo code are in:
frouingroup.ucsd.edu:/usr4/kyozo/calcite_albedo_CCM3_sw3/
The directory names such as EPS2_NH_SH/, EPS_NH_SH/, EPS_ri_sw3/, ttest/
are just for my convenience. These figs are just the 'draft' versions and
need to be modified for publication.
Also the CCM3 codes are in there that describe the parameters such as FSNS 
and how they are calculated.
The list of the parameters is in my_bldfld.F90, CCM3_output_params.txt, 
params_def.txt.

The figs of CLDTOT, vetically-integrated total cloud are in EPS2_NH_SH/
and EPS_ri_sw3/. They show similar patterns as some of radiative fluxes
but still it's needed to explain how the cloud fraction is related abedo
anomalies due to coccoliths.

The zonally averaged meridional circulation corresponding Fig 3 in Shell
with stream lines of the whole meridional circulation from coccoliths case
(vert_circl_cocco-contrl_YZ_JAN_vect_ri_sw3_a.eps etc) are in EPS_ri_sw3/.
They show some enhancing/weaking of the meridional circulation cells
but are not so evident as in the Hadly cells in Fig 3 of Shell.
There is a nice illustrated description of the atmospheric meridional circulation 
cells at http://web.unbc.ca/~ytang/chapter7_atmos_climate.ppt.

I realise that the fields of monthly coccoliths albedo used in the experiments 
are not smooth near some coast lines and redoing the experiments with smoother
albedo should be done when the computer time is available. Also
including the 'slab ocean' may improve the results (but it may need
further CCM3 code modifications).

There may be other unplotted parameters that explain the causes and effects of
albedo anomalies, but figures in color should make it easy to identify them.