I see what you mean, but it still seems like cancer resistance is a bit "overengineered" relative to what you'd expect from selection pressures. If typical cancer onset were e.g. at age 50 rather than 70, I don't imagine that having much selective relevance in the human ancestral environment, as you'd have raised child-bearing children at that point (and you probably can barely carry your own weight at that point, metaphorically speaking).

OTOH, cancer rate is probably superlinear with respect to error rate, so a small loss of cell instruction robustness corresponds to a disproportionately shorter lifespan.