Array Split / Replace Subarrays

[Yamaeda]

I'd use #2, the Index and Replace as free nodes. It's less code, thus easier to read and very rarely do you need to micro manage the memory.

IPE is quite helpful if you e.g. have a big array of clusters with fairly big arrays, and want to read or modify it. This is not such a case.

[dr2]

I'd just benchmark the code.  If the second option (Index and Replace nodes) is fast enough for your application (given the HW it's running on), why not make the code as "easy" as possible?

-DR2