You figure out these things when you have more time than money or sense. I
might not make it to mars before someone with a budget though.

Actually it turns out the second version might have better isolation,
because it turns out coupling between the resistors and the switch terminals
is more unwanted than between the attenuator sections themselves. But it is
wise to remember that probably 20db at each switch is the most to be
expected because of the coupling within the switch. It is probably a good
thing to compartmentalize also, to realize a maximum combined attenuation
with all sections engaged. His worry seemed to be trying to keep 50 ohms
between sections when all switches are bypassed, but I doubt that there is
50 ohms through the switches themselves. Of course if any section is
engaged, the return loss would make it a moot point.

I guess in the end, the switch itself becomes the limiting factor in the
performance of the thing, as with any RF switching project. Hard to find
non-inductive resistors too but surface mount chips might be a good answer
to that.