On point 1: this is the same as any other technology. However both HDP and CLDR growth is driven primarily by net-new use cases coming into production rather than new customer acquisition, which tends to be a (very) small initial purchase. The idea that "customers are struggling to get into production" is becoming less and less true as each quarter passes, particularly given how easy both companies have made it to run on cloud platforms (CLDR's IPO made much hay of having 30%+ of customers running on cloud infrastructure vs 18% for data in general). More customers are getting more workloads into production all the time, and both companies sport well over 1,000 customers each, admittedly with some overlap - particularly in FS.
Point 2: I was talking about cost of sale for CLDR and HDP. Selling an RDBMS or cloud is pretty easy. Customers don't need much education before you can start making money from them. Selling big data stacks is hard. Customers need education, training and barely-profitable services to be successful. I'm in absolute agreement that big data platforms are both substantially cheaper to run and deliver orders of magnitude greater capability.
Point 3: MapRFS has proven to be an enormous disadvantage in reality. It has no functional advantages, and hasn't for well over three years. That leaves it as what it is - a niche, storage-centric product onto which Hadoop ecosystem clones have been bolted on. It costs MapR an enormous amount of cash to keep those cloned APIs up to date (something they barely manage) and it utterly muddies the water for customers as to which bits they can use - the open source stack or MapR's (allegedly) compatible proprietary clone. It's no doubt technically interesting, but that neither leads to success nor makes money. The immediate business disadvantage for MapR is the cost of supporting such an obscene number of products. They simply can't do it with the cash they have/used to have on hand, borne out in a raft of analyst reports and customer gossip.
As an aside, CLDR these days mostly market their proprietary software, of which they have plenty (labelled in the IPO pitch as "HOSS" - hybrid open source). Even HDP, as much as they may be loathe to admit it, sell plenty of it. Can't exactly do a git clone on Cloudbreak or AtScale now, can you?
For me the proof is in the pudding. MapR are nonexistant in EMEA, APJ or China. They're nonexistant on the cloud. You can't survive like that, and that's more than clear in their recent announcements. HDP have just wrapped up a huge investment in LLAP, CLDR in their own PaaS (following on rapidly from Kudu). MapR? A "Cloud Scale Data Store" built on "Converge-X Data Fabric". What does that mean? MapR FS with a new label, targeting the converged storage use case.
Best case for MapR would be acquisition, but that ship sailed years ago. Only option now is a mid-sized IPO and a pure focus on the storage use case, and hope AMEX don't stop being a customer before they turn a profit. They'd be truly fucked if they lost that little cashcow.