I've been working with a similar design problem. You can see that Q3 and Q4 have "multiple drill" via arrays inside their drain pads, and there's a little green halo around each one which demarcates the "void" inside the soldermask shape. This padstack has no soldermask layer attached to it; that layer is defined in the foodprint, instead, and because it's a negative layer... adding "voids" means you've added an area that will have soldermask epoxy in it. Double-negation FTW :p
This allows me a nice seque into a related question: If you look at the above picture, and in particular those 15-via patterns in the drain pads of those FETs; it becomes apparent that there are no voids being created in the GND layer (the tan-colored background), unlike the "real" via which can be seen attached to a trace right above Q3. If these holes are through-plated they will result in shorts between the pad and any/all plane layers they intersect.
If this is compared to U6 (thermally-enhanced TSSOP14), the big difference is that the padstack is defined as a rectangle on every layer, whereas in the FET footprint the padstack is defined as single layer. Now -- I know that the fact that it has vias means that it's by definition not single-layer... but I don't know how to have no padstack (*except* the plated holes w/ clearances) on the layers in which I need no connections. The issue with what's been done to U6 is that the rectangle gets connected to the internal planes with a thermal relief, if I do the multi- rather than single-layer footprints. Kind of defeats the purpose ;)
If you look closely at the yellowish-tan GND layer at U6, there's a funny halo also visible there around the thermal pad. That's a shape that's been drawn, slightly larger than the thermal relief, with it's definition as "full connect". This allowed me to have through-hole parts still getting thermal relief connections, but these thermal pads to have nice beefy copper plane connections.
If the (top copper and top soldermask) Gerber layers are brought into GCPrevue; the result is this:
Which I think achieves the topside tenting thing you were looking for. Sorry -- but I can't find an option in GCPrevue to make the layers semi-transparent. That *might* made the above image a bit nicer to look at... =/
I'd really appreciate some comments/thoughts on this. I hope it might produce more interesting discussion on the tenting, thing, too...