The database you describe used to be called Pick. I worked full-time on pick from 1986 to 1990. It was written in the late '60's by the army and was WAY fantastically advanced for its time. I've seen an office with 12 people, 4 printers, and a streaming tape backup all running simultaneously on a 286 with fine performance and response times. Their "mulridimensional" database was amazingly flexible.
I really love this bit: Since computers weren't so great in 1969, they wrote a kernel that just emulates a better computer, then they wrote the actual OS on that virtual machine. When RISC became popular in the '80', the virtual processor in the pick machine was so close to the real RISC chips that pick was the first OS ported to the new IBM RISC servers, even before any IBM OS's like AIX.
I Love PICK. There's many variations these days (PICK, UniData, Universe, Ultimate, PICK OA, R83, R9, Advanced PICK, D3, MvEnterprise, Prime Information, Revelation, Mentor, jBase, Sequoia), many of whom started in the '70's. And Yes, I think Pick-like databases can run rings around any other db. They aren't just a db that runs on some OS; Pick IS the OS (tho it can be a guest too) IBM actually came to Pick first for an OS for the original PC, but its requirements were too much for the hardware, so Bill Gates won out. The owner of Pick actually laughed at the IBM guys; I'll bet he choked on that memory quite a few times.
Check out Pick - it really is what you describe and a LOT more.