I'm not sure what you're measuring... or what your -2dB is indicated on, so it
might be worth confirming a few things:
http://www.apicsllc.com/apics/Misc/filter2.html ... go to the bottom and plug in
driver impedance (in this case, use "8", as that's the L-pad you're using) and
"2" as the attenuation; with the l-pad knob set to "-2" you should be able to
measure R1 between the Lpad input and speaker output, and R2 between the
speaker output and ground/common, with nothing else attached.
...but the point of an L-pad is to approximate a consistent load as viewed by
the circuit upstream, while being able to vary the attenuation as seen at the output.
If you use an L-pad that isn't roughly close to the speaker load (or speaker plus
resistor load), the impedance seen by the upstream circuit (L-pad plus speaker)
will change with attenuation, and in a passive crossover, also affect the frequency
response. Dropping the 20 ohm resistor and finding a single optimized setting
for the 8ohm L-pad (attenuation - and parallel resistance) was just lucky.
If the upstream circuit is expecting the parallel load of 20ohms plus driver
(or a nominal 7.5 ohms if you make the -assumption- of a 12ohm driver), then
inserting an 8ohm L-pad seems appropriate. In setting up the 3133A network
though, JBL decided to use a 16ohm L-pad, so perhaps the 12ohm assumption
isn't the best. Do you have an impedance plot of an LE85/2420 on the horn
you're using? If not, I think I can set one up at home this weekend (no lens though).