Bug was reported by Miguel Nunes on the mailing list:

I found a case of memory leak in the standard 4 windows display.
I have my own window where I render my datasets, in my own plugin, and
when I interact with my window, parallel interaction happens in the
standard display.

after closing the window of the displays, no more memory leaks were
detected, so the problem does not come from my plugin or my own render
window.

This might be because my interaction takes the standard 4 windows to a
3D space where nothing is show, so maybe MITK is creating a valid 3d
space volume to allow the slices to go over there (and natural sample no
colours). If this is true, I might not consider it a memory leak, but
using some bounds could be a good idea.

I am using MITK 2013. 09, win7 64bits.

Best,
Miguel