I think the problem he points out is that the measurement (e.g.the polarizer/slit/field) is never perfect. In fact the measurements are also made of materials a

I think the problem he points out is that the measurement (e.g.the polarizer/slit/field) is never perfect. In fact the measurements are also made of materials at finite temperature with quantum effects so the subsequent probabilistic results won't be "exact". Of course you can design the measurements to be exponentially insensitive to those variations. However, when you build a sequence of measurements each exponentially sensitive to that variation you make setting up each individual measurement ever more difficult. It still doesn't prevent the thought experiment.