The main benefit of hisat2 is that is uses fewer resources than STAR and that it can better handler known SNPs if you make the aligner aware of them. Aside from that, I essentially always get better results from STAR, which is why we use it in our standard pipelines instead of hisat2 (this also bears out in published comparisons).

I have made the same principal experience as Devon. In addition, STAR seems to be much better suited to draft genomes and low quality genomes. The genome we are working with, for example, is of moderate size but consists of 33000 scaffolds, and contains ambiguity symbols. With STAR we on average achieve (unique) mapping rates > (90) 95% (<1% mismatches) and mapping speed of up to 400Mreads/h. With other aligners, which is mostly used on desktop computers, mapping rates are reported as low as 50%. Therefore, we have given up on Hisat, tophat & co.