I've followed Hail and applaud the Broad Institute's work wrt establishing better bioinformatics software and toolkits so I hope this doesn't come as rude, but I can't imagine an instance in a real industry or academic workflow where you need 300ms feedback from an experiment to "maintain flow" considering how long experiments on data that large (especially exome sequencing!) take overall? My (likely lacking) imagination aside I guess what I'm really saying is that I don't know what's preventing the usecase you've described from being performed locally considering there'd be even less latency?