I'm definitely a Blender newbie, but I've been working with models at Blender's default primitive scale, i.e. a new circle, cylinder or cube scale.
I build the model, export to STL, then check the integrity and dimensions with netfabb (the free edition). I can upload to Shapeways and check the dimensions of the bounding box - mostly it's height I'm looking at.
If the dimension isn't what I expect, I take a calculator and divide the Shapeways dimension into the desired dimension to create a scaling factor. This can be applied to the model in Blender then the model exported again to STL.
I know it sounds terribly fiddly, but I find the cameras/viewports work best at Blender default scales (I get a lot of Z clipping if the model is 'big') - maybe I need to practice with scene scales to smooth the workflow!
The opposite is true of Sketchup, where it's better to work in a huge scale, then scale the models down. I guess I'm applying that in reverse to Blender.