Crystals are cheap, engineering time isn't.
In the design phase while things are controlled by ET's, I would agree. But out on the assembly line, over time, production runs may not be so well controlled - especially, when demand outstrips the first batch and reordering for run #n is occurs. The cost for rework and later any warranty returns become significant product overhead.
Snippets from AN2606;
As a practical note, it should be firmly understood that 32768 Hz crystals are neither created equal, nor
close. Crystal parameters vary widely between different models produced by the same manufacturer, or
between manufacturers and styles of crystals. Designers must be aware that crystal oscillator circuits that
require ?tweaking? or that must have very high tolerance components to make the oscillator work are
going to be marginal, at best. Most oscillators work reliably with components that vary over a reasonable
range of values. If a crystal is used that requires high-precision components, the circuit designer should
give very special consideration to using a different brand or style of crystal.
Low-frequency tuning elements are more sensitive to temperature than their high-frequency counterparts.
In general, low-frequency crystals can be damaged more easily than high-frequency elements in soldering
processes. Particular attention must be paid to keeping soldering operations within temperature ranges that
will not damage the crystal. While high-temperature damage can manifest itself in many ways, the most
common symptom is the failure of the crystal to operate. Another way for temperature damage to express
itself is for the crystal to display bizarre or unexplainable start-up characteristics. If problems with a crystal
suddenly appear in a product line that previously had no problems, assembly soldering temperatures are a
good place to investigate.