The only purpose for the bearing retainer cage is to keep the spacing of the balls, and you could possibly argue that it makes servicing easier. More balls spread the surface load over more points, but more surface points could increase the surface friction.
Top of the line, groups from Shimano Dura Ace and Campagnolo Super Record used retainers in the bottom brackets and loose balls in the hubs and pedals. Running loose balls versus a retainer is only one or two extra balls different depending on the diameter. I think we are splitting hairs. Not only could you not feel the difference, but I'm also not even sure how you would measure it without fabricating some high-priced gauges.
I think the point we are missing in this discussion is that the driver side retainer is stationary (as in zero friction, zero movement) when the pedals and rear wheel are turning. The inside driver hub retainer only turns during coasting. You actually have only two retainers moving while you are pedaling, the left side hub retainer, and the right-side adjusting cone retainer.
The highest potential to reduce internal hub friction is by looking at the areas of highest friction/drag. Brake shoes and Brake Disk's have lots of drag especially if they were greased. A machined brake shoe cartridge has less drag. The second highest area is in the way the manufacturer designed the clutch assembly, and the retarder spring.
F&S (Sachs) and Perry are both great coaster brake designs. But in my opinion, the ramped roller bearing clutch driver, and the low drag of the lightweight retarder spring assembly of the English made Perry was never topped. If it were made today, I'm certain the cost to manufacture would be very high.
Perry for the win.
John