When I work on BGE projects, I always use development builds. This allows me to add new features to the engine and keep up on bugs. Also, if something breaking comes along, I can usually track it down to the exact commit and then I know exactly what changed/needs to be updated. However, this work flow is probably rare.
In general, we try not to introduce breaking changes, and when we do, we document them in the release notes. Any breaking changes that’s not noted in the release notes is most likely a bug, which, unfortunately, seems to be happening more frequently than I like.
Whenever I see in the Bug tracker that a bug is fixed, I update Blender. While using it I maybe bump into new bugs, which I’ll report then. Not really the right thing to do if you’re working on a project, because it may lead to unexpected surprises. But if you’re still learning and playing with possibilities, it may give you some inspiration. Another advantage is so you can help find the most recent bugs that entered.
It’s a first, but for now I’ll have to wait for some important fixes to be done before I’m optimistic about a future project again.
I think what needs to happen is that the near future should see less emphasis on new features and more emphasis on ensuring the ones we have work without glitches.
For example, issues that prevent libLoad(), libNew(), and libFree() from working optimally. I would also include fixing the world API for GLSL mode which to my knowledge is something HG1 worked on, and I would then include the effort started by LordLoki to organize the BGE codebase as that may serve as a good aid in bugfixing.
I update unless there is a specific problem with the update. Currently Blender 2.71 has a bug that 2.70 doesn’t that is stopping me from updating my game, but if it weren’t there I’d be using 2.71, definitely. Here is the bug if you guys want to see: https://developer.blender.org/T41813