An academic group ran similar experiments on Ion Torrent PGM, Illumina MiSeq, and 454 GS Jr DNA sequencers. And the winner was……….well, there was no clear winner - performance was differentiated by machine - the PGM was the highest throughput, MiSeq most accurate, and the 454 had the longest reads.
I think this is a problem because currently the incremental customers that represent sequencing moving from a niche to a mainstream activity are unlikely to have a full understanding of their needs. For example, while the team at the Broad Institute knows why they'd prefer machine 'A' over machine 'B', a typical pathology lab does not know enough to decide if they would get the most benefits from hardware producing the greatest accuracy or the longest reads? What's acceptable accuracy? I'd argue that the path lab at the University of Whatever can't answer these questions.
Until the answers here are more obvious, customer demand will be for a slice of a shared sequencing resource (a sequencing core) rather than for their own sequencing hardware, thus limiting the growth of the hardware market. (Though this is good news for Oxford Nanopore and other NGS hardware suppliers - the longer it takes for the market to mature, the more prospects who remain uncommitted to any certain hardware platform.)