The histograms in the top row are linear (normal) displays, expanded vertically in the middle row, and expanded further in the bottom row. The ideal thermal signal histogram will show a spike at zero with no or little tail towards the right. Pretty obvious that the 500D is far less impacted by thermal signal than the other two DSLRs. Specifically, even at maximum expansion, we note very few pixels that are so hot as to have reached saturation. If you wish to shoot 20+minute subexposures at an ambient temperature >25deg C, using an f10 SCT with narrowband filters, at a dark site, I'd suggest that you switch to a chilled astroCCD or a chiller for your DSLR. For the more typical rest of us struggling with achieving a decent, well-tracked, yield on 5-minute subs, I think any of these 3 cameras have noise that is low enough that it gets processed out easily. Specifically, a honking huge skyfog peak on the top left 20D histogram, at around 20 or 30% of the histogram X-axis, would make the thermal signal look trivially small. Just use MinMaxExcludedAverage for the stacking, and be happy.
For other stuff peruse my primary website: http://samirkharusi.net/