I think it will be impossible for us to asses how much it actually impacts function in real world use case.
Does seem fair though to say that if you have 85% less data input/probes, that you're losing some to a large amount of fidelity, than an algorithm can only make up so much for.
A potentionally bad analogy, but think of it as a high bitrate versus a low bitrate, for listening to music. The quality of the music will be notably different, but you would still be able to hear both of the songs in their entirety.
At the end of the day, it's a lack of data that was originally expected for the algorithm to work with, that is now missing.