In the introduction to the 1-sided coms chapter you are providing an example
of quantum chemistry applications. If you are refering to electronic structure
computations, the description of the 1-sided coms usage in this class of
applications seems to be not very accurate.
1. applications need not only "some shared memory for shared tables"
but most of the application memory is used for maintaining
distributed arrays which are accessed with 1-sided communication operations
like put, get, accumulate, scatter, and gather.
2. accesses are not "random" but only hard to schedule staticaly due to
the presence of dynamic load balancing
3. applications do not explicitly group application processes into
"memory servers" and the others (although an implementation of 1-sided
coms might be based on the "hidden agent" concept).
Please let me know if you need more information or references.