Couchbase in embedded devices environment

I would like to check the feasibility of using Couchbase in a highly distributed, embedded devices environment.

Suppose you have up to 2000 devices that come up and down on relatively long time line (days or weeks).
They need to share relatively small amount of data - at the level of may be 100MB for the entire cluster.
Each device only needs to update or fetch a document once in a few seconds.

Couchbase seems to support cluster changes pretty well, no need for master designation to any of the devices etc.

However, embedded devices are very tight on resources - a few tens of megabytes of DRAM, a few hundred of Mhz of CPU cycles.

Would Couchbase be able to operate in such a restrictive environment?
Would it be able to cope with 2000 nodes in the cluster, without constant chattiness between the nodes, claiming valuable CPU cycles for this purpose?
What would be the memory overhead i.e. how much would Couchbase claim on each node of such a cluster before the first document is actually inserted?

I’ve replied in a linked topic, as it seemed more appropriate for the Couchbase Lite category.