MapReduce is term pertaining to a software data retrieval architecture/process (also known as divide-and-conquer). The simple version is that instead of asking one super big database that knows "everything" you ask multiple smaller databases the same question i.e. "what all posts do you have from bob@domain.com?" (this is "mapping" a query to mutliple sources) and each database returns 0 or more results, then the query interface joins the results together ("reduce") to a single response. This is common in "big data" because you can more efficiently optimize the query by parallelizing it across many machines/workers/nodes. There are additional optimizations that can be implemented such as caching common queries or data-sharding (items a-f on node 1, items g - k on node 2...).

I don't think Nostr protocol is immune to the development of big centralized popular instances. Especially if something like Threads integrates and becomes the "default" client with millions of users over night. Users, in general, will always gravitate towards content and community. But, I think Nostr has a slight edge over ActivityPub in handling that problem by the user having no direct dependence on any one particular host.

I'll have to read more into the Nostr protocol specifically as it pertains to privacy, tracking and content injection (ads).

I'm by no means an authority on ActivityPub nor Nostr, I apologize if that may have been surmised. I too am just chatting.