[1] https://en.wikipedia.org/wiki/File:Conner_Peripherals_%22Chi...
(Decreasing IOPS/TB is also why SATA SSDs haven't gone beyond 8TB while NVMe and SAS SSDs are pushing past 30TB.)
I use HDD for bulk storage and SSD for high-speed storage. I don't mind less speed, except insofar as it impacts reliability (e.g. a drive in a RAID fails, and the redundant drive fails while rebuilding the array).
It seems like competitors to HDD have become increasingly noncompetitive over time. Tape drives are insanely expensive at reasonable capacity. DVDs hold 0.1% of an HDD and are more expensive. Various newfangled optical drives (e.g. M DISC) appear more reliable than HDDs, but also cost much more per TB. Plus, you need to keep a big pile of coasters.
I don't think that will be true. By having the heads only read half the platters, the bytes per track are <areal density increase>/2. That's more seek operations for linear reads and small offsets. For completely random access that's probably the same number of seek operations as before, but queuing latency is lower.
Where it might reduce seeks is where multiple processes are trying to stream data at the same time, which sounds like is happening more and more.
Their goal is to retain that TCO edge by ensuring that HD performance doesn't get too much worse as HD capacity increases.
I could imagine certain resonances (eg. the height of the head oscillating) might differ between the two heads, effectively meaning data written with one set of resonances cannot be read back with another.
IIRC the next step is to put the logic for that in the head itself with a tiny processor so the feedback cycle can be nearly instant. I suspect that would eliminate issues with one head reading something written by another.