So after playing with Tim's solution for a few minutes, it made me think of a different approach that ended up working perfectly for me (and quickly for a 600-parcel subdivision): use the parcel adjustment "against" itself! Here's what I did.
1. Create one or more "theoretical" control points for the plan based on the measurements and the POB. That is, these are control points that have nothing to do with the real world, but have a location calculated purely from the COGO measurements. For me, that was easy because it was a perfect N-S-E-W grids with all the blocks and streets the same size. For a more complex subdivision, you would have to create a fake parcel traverse to get out to the control points. I had really good results with the POB+one other control point, perfect results with the POB+4.
2. Activate these controls and deactivate all the "real" control points.
3. Select all the parcels in the plan and run the adjustment. BTW, one thing that is not documented well is because the adjustment is incremental, you don't get the best match the first time. You have to do a few cycles of "Run" then "Accept," until the Maximum Shift (in the upper right corner of the Adjustment Report) gets very low. In my case, about 6-7 cycles brought it down to 0 shift with 0 RMSE.
The result may not be exactly what was originally entered in the COGO measurements, but it is closer than you can measure. The advantage of this approach is that I can do normal parcel adjustments with the regular control points, and if it ever gets messed up, I can always bring it back and start over.