The question of rather to group or join the join the objects is a question function.

Rather its nurbs, polys, or subD's it still works out the same.. except nurbs can be a little different because you can patch model with them.

first if you don't plan to animate objects moving together in some way.. and you don't need to move or scale them all together in your seen file, you don't even have to group them.

like if you will not see the seam and there is not any real deformations accruing there that would make a seam visible thin its probably better just to group. With the little spider you would not see the seams because the angels where the peaces connect are very oblique and you have the geo pushed inside it.

mechanical peaces or architecture you normally do not want to deform. They just get translated, move, scale rotate, point constraints.. etc. or they just stay put. so toughs would be best to group. (note, visible edges of geometry can cause dark spots in MentalRay with GI and/or FG running.)

but for something like a head being separate from the body, its better to just join them in to one object, unless you have something big to cover up the seam. The reasons there, is that even if you over lap the 2 objects some. it will be near impossible to get them to deform exactly the same on both objects. Also 2 surfaces that are very close in face angels and are real close to each other will cause pixilation and or popping. The rendering engine and openGL get confused about what is in front of the other. BTW if your just making a still shot you might can get away with this..