Great thread - I'm about to do the same - subwoofer-only replacement.
There is a lot of good info on upgrading your sound here: http://www.dziedzic.us/wpc_speaker_replacement.html and here: http://www.flippers.be/pinball_sound_upgrade.html
The last link is interesting because it shows a more elegant way to control the cabinet/back box mix than an L-pad: a small potentiometer across R17. This is a great solution because it's potentially no soldering (I'll probably still solder mine), and the power amp no longer has to produce additional power (which can increase distortion and will very slightly increase your machine's power consumption), only to lose it in the L-pad. This fix reduces the volume going to the power amp, so it doesn't have to work as hard as it would with an L-pad. It's a pretty minor thing - if you are already using an L-pad I wouldn't worry about changing it.
I looked at the schematics and saw that the R17 hack will work fine. However it will slightly change the frequency response of the last stage (U1C). U1C + R17 + R11 + C7 form an inverting low pass filter. From DC to about 15kHz, the gain is -1. Above 15kHz, the gain starts rolling off to ensure the signal doesn't have any artifacts from the digital-to-analog conversion done to generate the signal (read more here http://en.wikipedia.org/wiki/Reconstruction_filter). If you decrease R17, that reconstruction filter won't work as well. The WPC-95 R17 link says that a value of 9k across the existing 22k R17 seems to work well. That means the effective resistance of R17 = 1/(1/22k + 1/9k) = 6.6k. So if you do that, the new corner for the low pass filter will be 53kHz (from 15kHz). You probably won't hear a difference, but it's possible high-frequency noise from that signal could cause audible distortion.
So another (and in my opinion the best) way to change the mix between the cabinet and the back box would be to remove one leg of R23 and insert a 200k or 250k potentiometer in series with R23 (between the free leg of R23 and the PCB hole where that leg of R23 used to go. This will attenuate the signal without changing the characteristics of the reconstruction filter.
Again, any of the the different methods of changing the mix (L-pad, R17, or R23 mods) will probably sound identical in the real world, but the R23 mod is the best in theoretical terms, so if you're making a mod that's the one I'd recommend.
I've attached a schematic of the WPC-95 audio filter stage (with corrections - the schematic had the + and - of the opamps backwards). I'll add pictures to this thread when I make my mod...