Loose balls may help, the number you use needs to be different from the number with retainer. The need for precision fit (Grade 25 versus 100) may be debateable, but for the moderate cost of 25's, and the need for good diagnostics (Is this installation REALLY parallel?), and the $40 minimum cost of a new HS, I'd go for the better balls. You've already faced and line-bored the head tube, so the bearing circles should be actually perpendicular to the axis and hence, perfectly parallel without added ride loads. This means you can torque the headset based on feel, which will equalize ball pressure as much as can be done (flex over bumps and landings cannot be eliminated). Conventional wisdom is to tighten ONLY as much as will eliminate play, but I think with a perfect installation it can be tighter. If it's tighter, it would take more ride load (forces of weight combined with impact loads) to cause the balls/races to separate due to head tube flexing. And it the grease layers have high integrity and the bearing is free of dirt (overhauled with enough frequency and perfectly clean), the good bearing surfaces should not wear visibly for a long time.

What I don't know is, how do you tell what the maximum allowable headset drag is? But it's not a wheel bearing or a BB.

I'd also take the bottom fixed cup loose, rotate it 90 degrees, and re-press it. One final thing is when it's apart, give the lower fixed cup and crown race a real good visual (bright light and a magnifier) and tactile (point of a ball pen or similar probe) inspection to verify the surfaces are perfect. If they're not, ball bearing tricks may help for a little while, but I'd think the case-hardened layers of the races and cones are already failing. In that case, the bearing surfaces are at the end.