...couldn't I use a retarding film of randomly varying thickness to convert a Coherent laser beam into an incoherent laser beam to improve eye safety?
Absolutely not. A beam's destructiveness to the eye depends on three things:
- Energy delivered to retina and the time periods it is delivered over, quantified by the ISO60825 concept of Maximum Permissable Exposure;
- Spectral content: some wavelengths are short enough to cause photochemical damage: Maximum Permissable Exposure limits in ISO60825 are wavelength dependent;
- Indirectly, in some cases, whether the beam can be seen and thus whether the viewer can thus shield their eyes through the blink reflex. This is the grounds of the laser classes 2 and 3R: a fleeting viewing of the former cut short by the normal blink reflex is deemed safe by ISO60825 and the same glimpse of a 3R beam is of greatly mitigated risk.
Coherence has nothing to do directly with a light beam's hazard to the eyes. Indeed the laser safety standard ISO60825-2014 makes no distinction between coherent and incoherent sources. You assess the safety of an LED in exactly the same way as you do a laser. A proviso to this statement: if you add aberration or incoherence to a beam in the way you propose, you will change its divergence / convergence behaviour and thus you will change the power density reaching the retina in certain situations. Some coherence properties thus influence, indirectly, the calculation of the Nominal Occular Hazard Zone and Nominal Optical Hazard Distance from sources.
Moreover, to simulate incoherent light as in Punk Physicist's answer, you need to answer exactly what you are simulating incoherence for. To be pedantic, a truly partially coherent beam is a classical mixture of pure photon states, as I describe in my answer here. So, the only way to produce truly incoherent light from coherent light is to randomize each photon's phase / polarization by making it interact somehow with a truly random process: a thermalized system or radioactive source, for example.
The methods for decohering light in Punk Physicist's answer only produce pseudo-decohered light, which has the same relationship with a truly incoherent / partially coherent source as a pseudo-random number sequence produced by a deterministic numerical algorithm has to a a truly random sequence produced by e.g. a quantum system. This is nonetheless useful for many applications, in the same way that pseudo-random sequences are highly useful.