It's useful on all kinds of transactional/bigger data, IMO (up to ~1e9 rows). It's also extremely useful as a prototyping language for complex mathematical ideas.
Just in case, a good implementation (e.g. GNU APL, Dyalog) to recommend (performance-wise)? I also tried J years ago, but not on larger data sets or computationally expensive things. My experience and overview is pretty dusty.