One reason Brain is (probably) slower on CIs and example builds is that it does at least some unnecessary updates. For example on #8985 + #8975 if I add a mne.set_log_level('debug') in some test script, each time I tweak a clim or update a time point I get:

>>> brain.set_time_point(0)
Color mapping 'curv' with Greys colormap and range [-1, 2]
Color mapping 'data' with <class 'numpy.ndarray'> colormap and range [4.73, 14.18]
Color mapping 'curv' with Greys colormap and range [-1, 2]
Color mapping 'data' with <class 'numpy.ndarray'> colormap and range [4.73, 14.18]

So it's being mapped twice. And if I modify the _Brain code to have this:

    def _update(self):
        logger.debug('Update')

Just on init I see:

Update
Update
Update
Update
Update
Update

So we get 6 renderer updates. This last part might not actually matter since just having _update return immediately it's still ~4 seconds to open the window and exit.

So this is mostly an issue to remind us to track down and profile to see where time is being taken.