I suppose the idea of reducing freeze time with "always releasable testing" didn't really work out (lack of resources?).
Unstable is almost always a better choice. Things may occasionally break, but fixes will arrive very quickly.
Unstable is the Debian branch to use if stable is too slow for one's tastes. The testing branch is just for testing and it's really not safe to use on the internet.
Can't you make the same argument against using stable for the same reason?
> If something breaks, it will stay broken for quite a long time until the fix makes it out of unstable [and testing].
This won't happen with stable, since it will get the consistent result in the end.
on stable you don't get data corruption bugs, and you shutdown or filter security holes until they are confirmed fixed.
on testing/unstable you can have data corruption bugs and everything else. but you may bet that security fixes are done faster. emphasis on bet.
Basically you install the Stable system to get a stable system, and then 'bring forward' just those packages you need.
I was talking about the freeze itself though. It affects testing too which is normally rolling. So if the freeze is too long, things start becoming annoying.