It's likely the repetition rate that caused the failure, with the designers expecting enough time for the snubber device to cool between switching cycles.
With PWM drive, the snubber dissipation would be far higher and it sounds like heat just accumulated until it failed.
A flywheel diode across the coil should solve it, as you say.
We manufacture some little power switch modules for controlling clutch-brake units with a typical rate from 0.5Hz to possibly 3Hz at most, and the snubbers (disc VDRs) are by far the largest components in them.
There is no way I'd consider PWM drive to an inductive load with anything other than flywheel diodes.