Not an IDDQ expert, but if the voltage changes then the leakage current through the part would be different (V=IR). If the IDDQ test is automated, then perhaps the acceptable limits to detect problems would no longer be valid, leading to either undetected problems or non-problems being flagged.
Certainly every device's leakage will change, and a
"weak sister" may be exposed by higher voltage that
was "in the background" at lower. And dynamic current
goes, at minimum, as CVf.