BUG: Slicing timeseries with over 1000000 entries with time fails

Slicing a timeseries with time results in a TypeError if the timeseries contains over 1,000,000 entries.  

```
import pandas as pd
from datetime import time

ser = pd.Series(index=pd.date_range('2000-1-1', periods=1000000, freq=pd.datetools.offsets.Second()))
ser.head(999999)[time(15,0)] # this is fine
ser[time(15,0)] # this fails
```
