I suspect that the paying users are more likely to listen to more obscure indie artists, simply based on the people I know who pay for premium. Anyone interested in listening to only top 40 likely won't see the value in paying $9.99 a month for it. Thus the less popular artists are bringing in the revenue but the big artists keep more money.
That being said, there's no proof of this without data to back it up.
If I were to speculate, I would expect the number of times a song is played most likely to adhere to a power law and therefor this average number of streams is not an ideal measure.
And yes, I'd would expect the tail to be fragmented. Lots of people listening to but a small and mutual exclusive set of bands. Next to the odd main stream one.
So perhaps it is not really a question of light users sponsoring heavy ones, but more the banality of average taste.
However, he then makes the leap that that means that less popular artists are subsidising the popular ones, which has yet to be proved.