I’m pretty sure I know how virtual heightfield meshes work, Unity is doing the same thing and in fact sponsored the original research. It’s fully scaleable to large sizes and should have a relatively constant runtime cost, just like nanite. There are a handful of scale considerations to it, but nothing an ok-ish sized open world game would be bothered by, and those problems could be overcome with some effort to get to nigh arbitrarily large scales.
It’s also possible to apply this to arbitrary meshes. It’s just a software version of tessellation that’s crack free and has a fairly smooth and continuous transition. Long term, if Epic dedicates the resources to it, it could be a full replacement for tessellation, including programmable materials and displacement maps and etc. It can even do catmull-clark subdivision, though Epic has stated they don’t want to work with non standard modelling tools like that.