If they didn't care about using the same ISA throughout, there are plenty of tiny free low performance RISC-V cores you can just grab off github. Ok, not *as* small as an 8051, but also not as low performance. Comparable to a Cortex M0, in fact :-)
Hehe, would probably be much better here, but I bet this would still cost more for them. Implementing a "random" free core on github is not 100% failsafe and takes some work, with potentially no support. They likely had everything ready for the 8051, including possibly the layout (if, as suggested above, it's actually not all cores on the same die...)
And yeah, the 926 is oldish now and probably doesn't cost too much. The expensive part is probably the quad-core Cortex A7. I didn't see the specs of the 926 they used; how fast can it be clocked? It doesn't seem very competitive compared to the associated quad-core A7! (But likely reason to integrate it is for the Linux support I guess?)
Which all in all, makes this chip look like just an opportunistic arrangement rather than innovation.