I think this proposal is problematic because it allows a heterogeneous distribution of block data across clients. The rational node operator will only maintain the minimum amount of historical data. Therefore, we can expect the vast majority of the network to not provide any substantial amount of the history. To find nodes that do will require a long period of searching the discovery network and then the nodes that do serve most or all of the history will be fully saturated by those looking for it.
This is a solution that I feel avoids the problem of making all historical data highly accessible to nodes. My proposed solution has been to implement the era and era1 format in ELs and bring many different history distribution mechanisms online (e.g. bittorent, direct download, portal network, etc), while retaining the property that all nodes on the network have the same historical data, albeit a smaller amount, such as only 1 year or 3 months worth.