Frustrating this is. I keep coming back to Gary Honis' Cooler box design Mark III which yielded a 43.7F (or 24.27C) drop in temperature. No cold finger. His Dark Frame tests show substantial improvement, but without a temp sensor near the sensor chip its hard to know what real drop was realised.
http://dslrmodifications.com/rebelmod450d16c.html
Page down on that link page to see his dark test results and images.
I noticed that there was another prototype that experimented with gas phase change cooling, where he was running gas into the sensor area and had constructed a little manifold cold finger.