This is what the innards of a variable L-pad look like:

What it does is present the amplifier or passive crossover (between terminals 3 and 1) with the same (-ish) impedance while attenuating the speaker.
Left most variabl;e resistor is 8Ohms, rightmost one is rather high (1k5? didn't do the math ;P ).
At it's max setting, the 8Ohm is out of circuit, and the speaker is parallelled with the high resistor. This yields something close enough to 8Ohms, and (almost) maximum power to the speaker.
At it's min setting, the speaker terminals are both shorted and the 8 Ohm resistor is in series, still presenting * Ohms to the amplifier.
That's one mono one.
If I'm not mistaking PRR's solution does yield 16Ohms but doesn't do variable attenuation - at both extremes the speaker is parallelled with the high resistor, and this combo is min series with the 8 Ohm resistor. Attenuation, just not variable.
I don't really see a way to do this with a stereo L-pad, though I might totally miss something.
That said - valve amps are quite sturdy as far as impedance mismatches are concerned - I wouldn't worry about a 1:2 mismatch myself. Just don't leave the speaker terminals *open*.
Using an 8 Ohm pad will result in presenting the amp with a just-under-16Ohms load at full tilt, and an * Ohm load at full-off. Don't see any reason why that wouldn't work, there's no crossover filters that could be thrown by this.