Just based on viewing implementations, I would say that bigarrays can be slower if you create a large number of short-lived arrays. It seems that the memory for them is managed outside the usual OCaml GC, which handles short-lived objects very well.
You may also find that calls to bigarrays are not included, while access to inline arrays will be.
On the other hand, inline arrays will have additional indirection for two dimensions.
If performance really matters, you may have to test your specific application.