In[34]: df.sort_values(['job','count'],ascending=False).groupby('job').head(3)
Out[35]:
count job source
4 7 sales E
2 6 sales C
1 4 sales B
5 5 market A
8 4 market D
6 3 market B