Jesse from VT recently sent me back some bench test results for the WRT54GL that finally confirm our assumptions about output power and the effect of power settings in OpenWRT.
According to ye olde spectrum analyzer, the output power is a smidge lower than the 84mW that is commonly stated on the interwebs. Jesse recorded 13-15 dBm over the width of ch1 after passing through 4 connectors and about a foot of cable. Assuming 1/2 dBm loss per connector and a touch more for the cable, output is roughly 18dBm for ch1. As expected, the power settings in OpenWRT had no effect on output. Although we didn't measure all the channels, Jesse suggests that the radio is likely a little stronger in the center range (maybe even as stong as the full 19dBm we'd expect).
Jesse further verified that the Power settings in OpenWRT had no effect on output.
Thanks to all this info we can now firmly state the gain on the large antenna is 13.5-14dBi.
[Edit: a picture says 1000 words. Here's the photo of the bench test setup that yielded that data above. As can be seen from the photo, the tested antenna is the weaker of the two. Jesse is hot on the trail of another test using the other antenna so we can compare.]