Thanks to everyone for the quick and very helpful replies. For future reference, I found the annotated schematic and pointer to a QST article analyzing the RS-35M: http://www.al7fs.us/astron-rs35m%20annotated%203.pdf
The actual problem was more mundane and really taught me something: take Ohm's law seriously.
When I went back to start the diagnositic process, I measured the no-load and full-load voltages right at the P.S. They were 13.8v NL, and 13.6V at 20A load. Hmmmm...
I had originally been measuring my voltages toward the rig end of the power distribution chain - which in my case included lengths of 10ga and 12ga wire, fuses, a RigRunner panel, and multiple Powerpole connectors. So the Astron was certainly fine, and the full-load voltage drop was occurring in the power cabling.
I shortened and simplified the cabling, which reduced the drop quite a bit. Even then, when I measured the actual drop in the cable with just ~5ft of 12ga wire and one set of powerpoles, at 20A load I had almost .2V drop PER SIDE, or ~.4V overall. Within measurement error this basically agrees with the standard resistance figures for 12ga copper wire.
Lesson learned: Even small series resistances accumulate in the power cable chain, and at high current can cause a surprising degree of voltage drop. Use big short power wires. (And don't blame the P/S).
This is one of the downsides of low-voltage/high-current arrangements that we have to live with.
- Les, W9XC