Profile test results:
====================

The following shows that the Workspace refresh method is responsible for the
observed lack of performance.

--light: 30ms
(same as IPython PyQt4 frontend which implementation is far more simple)

No option (or --numpy, or --scipy): 50-60ms

When Explorer widget is disabled (.spyder.ini): 30ms 

--all or --pylab  :

* Workspace auto-refresh is disabled: 50-60ms
* Workspace is empty: 140ms
* Workspace contains 2 numpy arrays (shape: (1, 2e5)):
    - workspace hidden : 140ms
    - workspace visible : 300ms -> adding more arrays: 80ms/array of this size
                                -> proportionnal to array size
                                -> show complete values: ~2-3ms/array!

==> Explorer contribution:  ~30ms -> solved by refreshing only when necessary
==> Workspace contribution: ~80ms when empty

-> when Workspace is hidden: the most time consuming task is filtering

=> Workspace lack of performance solved too! -> was due to a CONF.get within a
   loop... duh...