I don't know of any issue of high failure rates on PowerFlex drives, in fact they have arguably the lowest failure rate in the industry. Of the less than 1% that fail, roughly 80+% of field failures are attributable to heat related issues, 15% to power quality issues and 4% to application issues. The defect rate is a fraction of a percent.
So assuming you are comfortable with the cooling system, and the application is correct (per kwired's response), power quality is a likely candidate, especially if the caps are involved. On the large drives like that, the big caps have what are called "balancing resistors" across them, which help to even out the stresses the caps see from grid transients that come right through the bridge rectifier. Years of severe transients will take out those resistors, which is followed shortly thereafter by cap failure, followed shortly thereafter by transistor failure. Regular preventative maintenance programs will include inspection and periodic replacement of those balancing resistors. Ignoring PM on drives leaves those resistors (and other things) as ticking time bombs.
Usually when a cap fails first though, the drive will fault in high DC bus ripple to prevent transistor damage. That would lead me to suspect that the failure happened the other way around, that a transistor failed first. There are protection systems to help avoid cascading failures in that direction too, but often, without load reactors, a serious event on the load side happens too fast for the systems to kick in. Whenever people ask me if load reactors are necessary, I always explain that no, they are not. Neither is health insurance (can't say that any more though), but when it turns out you needed it, you will wish you had it. The main benefit of a load reactor is to slow down the rise time of a load side transient event, such as a motor winding or lead cable short, to where the drive can act in time to limit the damage.
You say you have 3 that failed? All at the same time? If so, that points back to a transient event on the line side again. It only takes one significant event to do damage across multiple units.
Another issue, what is your power grid like? Is it an ungrounded delta or a high resistance grounded wye? If so, there is a jumper on the incoming rectifier section that had to be removed to prevent MOV damage. Was that done? If it is a solidly grounded wye system, did someone inadvertently remove that jumper? Those Frame 8 drives usually also have common mode chokes that need the jumper removed under the same conditions, and people used to doing it on smaller drives that don't have common mode chokes sometimes miss it on the larger ones that do. Over time, that can lead to damage too.