Hi GHP,
I think you just need to not build models that rely on fusing of adjacent parts. If you want two meshes to print as a single object, you must explicitly overlap them in space.
That is, even though two parts are 0.3mm apart, which will cause them to print as a fused object
due to the statistics of the printer's resolution, SW will not allow you to claim that this printer-error-induced connection constitutes a real structure. (I hope you would agree -- what if the printer actually does what your model specifies, by chance! Yikes!)
When modeling for SW, I create virtually every part with a separate mesh. (Here I must respectfully offer the opposite advice from BillBedford.) In my
Scarab and
Scarab 2 models, there are more than 20 separate meshes, each carefully overlapped to create a continuous solid object. The downside of this is that I pay a bit more because overlapping material is double-counted by SW. Currently, they compute cost by summing the volumes of all meshes as an approximation to the true total material volume.
The upsides are: first, and most importantly, I can adjust each segment of the beetle independently just as if it were a real creature, which has helped enormously with the model refinement process, allowing me to quickly turn Scarab 1 into Scarab 2. Second, I'm lazy, and Boolean algorithms in Blender still frequently produce non-manifold problems that I hate debugging as much as anyone. No Booleaning at all in these models. (I did print signet rings for me and my girlfriend that used a rather elaborate Boolean difference operation to create the stamp portion. Worked fine, but it took forever to fix the non-manifold problems.)
So, my advice: act as though the printer will do exactly what you tell it to, and don't rely on its errors to produce solid models. Use explicit overlaps if you like, or unify your meshes. Both work. The former is more expensive and requires more thinking but is more flexible if you intend to make major changes to the model; the latter is maximally cost-efficient and foolproof but can somewhat inhibit editing later.