In ROOT 6.24, the BatchMode was still experimental. It didn’t give the right result for most fits and was also not necessarily faster.
Have you tried ROOT 6.28? HistFactory got a major speedup there (if you also generate your workspace with that release, not only fitting), and with the BatchMode it should be again around 30 % faster. ROOT 6.28 is also the first release where the BatchMode gives the right result for all usecases known to the developers.
So please try 6.28, and if the BatchMode is not faster there it is considered a bug. Then you should open a GitHub issue and provide a reproducer to we can debug this problem.
I see. Will try with ROOT 6.28 and report back. Just want to confirm: our usage of the BatchMode is correct (e.g. there’s no need to supply separate/additional compilation flag) as long as we update to ROOT 6.28?