Forget this. My test was flawed because I still wasn't comparing apples to apples. I was comparing the pre-bootloader state to the post-bootloader state, and it seems that OBP, even on a real machine, shows all the registers as zero before it runs any program.
However, I still think there's something wrong with the RT625 version somewhere that causes it to show as a 605e in the banner, but I don't have the real thing to compare against. What I do have for comparison is a 150MHz hyperSparc which shows up as an RT626. Bob