I think there might be a timing-related issue.
On my build, both the SSR (that turns the amp power on) and the electromechanical relay (that switches the bleeder resistors) are driven together by the ε24 power switch driver circuit. So they mostly switch simultaneously (the electromechanical one just a hair slower due to the physical movement, but then the SSR may also delay very slightly because it would switch only at the AC waveform's zero-crossing point). Since it takes a small amount of time for the transformer to "set up" its magnetic core and for the large amp caps to charge up, the bleeder resistors are switched out first.
On your build, your electromechanical relay is powered up by a σ25 off a different transformer secondary. This means that when the relay coil is activated, the transformer will have already done setting up its magnetic core and the σ25's bulk capacitor is already charged up. This also means that the big amp bulk caps are also already charging up, at this point there is probably a large current flowing through the bleeder resistors. The delay is still very short, but might be long enough to make a difference.
Now that's a total guess. Maybe your relay is wired wrong or otherwise damaged, I don't know. If you have a way to check the relative timing of the events (with a multi-channel digital storage scope running in single-event trigger mode, you could observe the rampup of the main amp rail voltages and the relay coil voltage to see this) it would bear out whether my guess is true. A simpler test is to measure the voltage across the bleeder resistor while powering up. I don't know whether the DMM will respond fast enough to register a reading, but you shouldn't see a big voltage spike.