Interesting concept indeed! Looks like the large outer rollers would probably have to be molded in two halves, then bonded together. Not impossible, but also not particularly cheap. The real trouble will be with the hollowed-out portion of the larger rollers, because any material soft enough to be grippy will also be soft enough to flex inward and rub against the support inside it. One might perhaps skirt around this issue if the surface it would rub against is replaced with a bearing, however, which would be even more interesting to see.
I'll take your current design, and raise you an improvement:
Most of your assembly difficulties could be eliminated if, instead of alternating between small and large rollers, you made all of the rollers the same size, but asymmetrical: small on one end, and large on the other end. The small end of each roller would nest inside the large end of the next roller, and so on around the circumference. That way, you only have to solve the interference problem for a single roller instead of all of them, and your production costs would be much lower.
Here's a quick mock-up cross-sectional view of what I'd suggest:
(it's missing the roller pins, but you get the point)

The roller profiles and support arms would definitely need some tweaking to make them more structurally sound, and you'd need to work out the assembly process a bit more, but it's not all that far-fetched to think that this sort of thing could work IRL.