And they are not an order of magnitude different in difficulty. Witness the success of PLoS and the Journal of Machine Learning and (to some extent) arXiv. The journals coordinate peer review and provide aggregation; now that we have the internet, there are better, much cheaper ways to do it.

Is there any reason to believe that this is the case?

I'm a member of the ACM and the IEEE, and a lot of other people are; it doesn't seem to me that the modest dues are significantly hindering progress in the field. If, due to some unexplained fortune, the ACM and IEEE were to drop their paywalls tomorrow, and grant free access to all, how much would really change?