I typically use 2% max variation from plumb.
This is subject to modification. Piles in groups can deviate quite a bit more, as long as the piles are not interfering with each other and the pile cap can perform properly.
Oftentimes the real issue is the eccentricity of reaction. If a pile is not plumb, it must be determined what is out, either the toe or the head. Is the pile a single or part of a group? The issue usually becomes a problem of statics.