The pattern-seeking brain would be driven crazy trying to predict when the next tick is going to happen, as this pattern is not easy to analyze without tools. Experienced musicians could figure out that the shortest time between beats is half the second-shortest, and perhaps figure it out from there.
Anyway, you could make a website that simulates this or generate a long YouTube video, send a link to unsuspecting people and see what they think. If you want to be extra sneaky, use rain sound as background and "close-up" recordings of single drops for the beats. If you can't code, make sound files of all the different possible measures in Audacity and use a media player with seamless playback and naïve shuffle.
This is similar to some popular exercises for improving your internal pulse. E.g. having the metronome drop out for a number of bars while you're playing.
My prediction:
On its own, it would be hard to derive the underlying pulse. Even a trained musician would take a little while (my guess is 4+ measures). In the context of a song it would probably have little to no effect.
Well, something like this is actually quite popular in modular synthesizers community. They have one type of modules called "Clock generators" which generate gate/trigger signals for given BPM (Like 1/4 or 1/8 or 1/16 rhythmic pulses for 120 BPM for example) and another type of modules called "Bernoulli gates", which basically allow to specify probability of input signal going to the output. Those beat-skipping metronomes with configured probabilities are then used to trigger notes or samples or whatever. Also, this is modular where you can modulate almost everything, including BPM itself, but that's a different story... Stochastic music approaches like this are often called "alleatoric music".