From fixing IBM mainframes in the late 80s, they were far more unreliable than modem machines. Or at least, a mainframe consisting of many interconnected frames joined with thousands of tiny cables could throw up wicked problems that required hours or days hunched over an oscilloscope or logic probe to isolate. They probably seemed more reliable because of relentless redundancy in the hardware and a highly focused maintenance staff who proactively followed up on logged intermittent errors before they became serious.