Workshop observation from cleaning a batch of three-year-old towers. Every one of them had a front intake sufficiently loaded with dust that the fan curve had shifted upward at idle.
I did not instrument this properly, so I cannot give you a delta. What I can say is that the machines were audibly quieter afterwards at the same workload, and the owners had all independently described them as "getting slower", which in at least two cases turned out to mean "getting louder and hotter".
If someone with a proper thermal logging setup wants to do this as a controlled experiment, filter loading seems like an easy and genuinely useful thing to quantify.