I did a little test to see whether small or large masks had any influence on render times and memory consumption. In the chart you'll see that it doesn't really. I would have thought a 1.5 Mb JPG would use less memory than a large greyscale TIFF, LZW compressed. At first I even used 40Mb RGB psd files (TG didn't load greyscale psd), but wrote them off since today.
I did this because I have to produce a large render (a real blow up of 2x2 meters), and had to include many roads. They were a bit soft in my first tests, so I wanted to use a 200% mask, and thus sharpen the outcome. As far as this test goes, it should work ok this way, with no setbacks in memory use and rendertime. The only positive thing of a compressed JPG is that it takes less space on the HD. The quality is quite the same as a full TIFF, I looked at this in detail.
Does anyone have similar or other experiences with small versus large masks? Are my assumptions right?