Doesn't terminate in a reasonable amount of time.
One observes that changing :gen-max doesn't affect computation time.
One observes that it can be fixed by adding :into  to the s/coll-of spec.
The reason is this: If :into is not provided, s/every and s/coll-of has to generate a vector (via gen of vector?) to call empty on it, to then fill it up. This is quite clearly documented in the docstring of `s/every`:
Assumedly the vector? generates quite large vectors at a certain point which significantly slows down the generation.
The responsible code is in gen* of every-impl