But the argument is that loaded friction is different than unloaded (I'm a sceptic
IMHO at the low range of revs most bicycle components that are running high friction unloaded is still going to be high friction loaded.
Loaded friction may be less (probably will be) than unloaded friction though if and when that friction is caused by too much lube or too tight initial tolerances of the bearing, for instance. As in no bedding in has occurred yet.
OTOH a bearing that spins freely unloaded may exhibit higher friction when loaded for the same reasons but in reverse. I.e a lack of lubrication and/or not so tight tolerances allow a bearing to spin freely unloaded.
When loaded however this bearing's low level of lubrification may now well increase friction as the contact area with the races increases significantly especially when the load is not a simple radial one but also a lateral one.
Ideally you'd like a cup and cone bearing that, no matter what load or what angle, is going to act as a single line contact bearing with an extremely thin contact point/line at any angle.
Deep groove bearings work in a similar manner, alternating conctact surface direction depending on radial load but when running perfectly horizontal the balls see much more contact area than a mere single line contact. They're surface area of contact is roughly twice when new but when used is approaching more and more the shape of the grooved race.
Hence the contact line is growing wider which invariably increases friction.