The problem is if they run (very) hot the memory might develop latent failure. Sadly this sort of failure causes crashes and corruption under stress (parity errors); and generally not our-right failure to run.
If a GPU/VRAM is failing it usually because:
1) It was originally defective from the factory
2) Electrostatic discharge
3) Capacitor failure (not usually an issue on the better brands)
4) Someone was REALLY overzealous when OCing and went as far to use overclocking and firmware tools that went beyond the voltage safeties.
Virtually every modern GPU is designed to throttle itself with a LARGE degree of tolerance. Even if the card was under full stress, the GPU would become irrelevant for games, long before it failed. Theory and practical application are vastly different.