I also found out about a new project at Uppsala University that is attempting a subset of the Cloudi functionality, but with a different aim. The Low Power Erlang-Based Cluster (LoPEC) project is focused on low power computation using OpenCL integration. So, it will be interesting to see how their end result might be influenced by Cloudi.
One point that might be an important distinction is that the LoPEC project seems to be concerned about internode communication of work data (based on conference discussions). The Cloudi framework enforces an interface that does not include internode communication (inter-work-task communication, i.e., internode in this context does not relate to Erlang nodes but rather computational nodes which are worker threads in Cloudi) so that work parallelism is preserved and maintained (at the cost of only supporting "embarrassingly parallel" and "divide and conquer" algorithms). The Cloudi approach is part of a hope that divide and conquer algorithms can be created for work (using Functional Decomposition) that would otherwise rely on simplistic internode communication that prevents parallelism. Such divide and conquer algorithms are not trivial, but should be a goal for efficiency. It is possible that MPI or a similar internode communication method could be used to communicate work data within Cloudi work but that direction would be part of a conscious decision to bypass the framework.