If you can get the bowling pins to stand up and you're just looking to find the place where they won't fall any, then the best option is to enable the debug drawer and take a look at where everything is. You can also do some calculations based on the dimensions of your loaded model and cylinder to get the precise location. For example, if you know your top triangles are at Y = 0, and your cylinder has a height of 10, then your cylinder center position will need to be around 5.04 to 5.08. This comes from (height / 2) + collision margin * 2 - some small epsilon.
The collision margin is explained here:
http://www.bepu-games.com/forums/viewto ... ?f=4&t=409
If you can't find a spot where they will stand up, chances are you are encountering some floating point precision issues. Cylinder-Triangle is handled by the general case collision detection system which is sensitive to such imprecision.
If your bowling surface is made up of two long, stretched out triangles, the interaction with the tall cylinders will likely be enough to introduce small errors which become visible as wobble or instability. Increasing the gravity to compensate for the scale-induced moon gravity will likely make the problem more visible.
When it comes to making robust collision meshes, it's a good idea to have fairly regular shaped triangles even if it means having a few extra triangles in the collision mesh. If the scales were smaller to begin with, this might not be necessary.
Robustness and speed of the general case collision system is something I'd like to work on in upcoming versions (collision revamps make up a good portion of what's on the version roadmap:
http://www.bepu-games.com/forums/viewto ... ?f=5&t=849). But there are some fundamental limits imposed by single precision floating point numbers and the algorithms involved, so these requirements, even if looser in later versions, will probably have to stick around.