Not sure. I actually don't use them. I put EVERYTHING in my glossary. I keep trying to use Non-translatables, but you can't catch any alternative spellings/synonyms, so I find it easier to not use them.
The biggest advantage of non-translatables is IMHO that they are not spell-checked on one side and on the other side they do not show up (= waste space and attention) in the (visible) glossary. Indeed they should respect word boundaries.
Another nice feature is that you can fill all segments that only containĀ non-translatables, like brand names and numbers.
@Torsten: that is the main reason I keep trying to use them: they do not show up (= waste space and attention) in the (visible) glossary. I am always investigating ways of pruning what is shown in the glossary pane.
@Hans: I'm afraid I have yet to encounter a situation in my texts where many segments contain only brand names. I do sometimes have a bunch of segments that contain only numbers. However, I haven't yet managed to make CT reliably auto-translate numbers (dates, currency, amounts of money, etc.) without having to check them all, one by one, myself anyway. No matter how much I fiddle with the various numbers settings under Edit > Options, some of them always get done wrong. That is, one type of them usually gets done right, but then another doesn't.
Hans CafeTran Wiki
Obviously I've entered one non-translatable 're' in the past. Now I get this orange colour coding:

What could be the disadvantage of making non-translatables respect word boundaries?