Agreed. They should have required an extra token like ? to disambiguate bindings from evaluations. Not only would it avert this pitfall, it would also obviate the need to have a special identifier to discard a bound value; instead of writing _ to discard a binding, one would just write ? without a variable name.
I fear a poor design like this is going to not only discourage people from using the feature at all, it is also going to poison the well and discourage people from using pattern matching in languages with a much more robust implementation. Which is a shame, because I actually like pattern matching.
(Tinfoil-hat take: Maybe that’s the whole point, after all Guido doesn’t like functional programming.)