Thanks Mossie
I'll scale everything up before doing a proper boolean merge if that's what it takes to avoid anomalies.
Regarding the question asked by "blackjak" I also wonder to what extent the clean-up matters in terms of computational cost / frame rates. Obviously, the higher the number of vertices / surfaces the higher the computational cost. However, a "clean" model may very well end up with more vertices than one where a cylinder is simply stuck into another cylinder. The proper matrix of surfaces requires new and possibly additional vertices, particularly when two "curved" models are properly joined. Is there a simple way to judge this? I'm thinking of number of surfaces / vertices (polygon count) or total file size.
In any event, a totally hidden surface should certainly be deleted since it wastes resources without contributing anything.