Inconsistency is a fact of life, and any repository instance or system that wants to avoid bottlenecks is going to have to accept items that have inconsistent metadata (and possibly inconsistent formats and policies, though consistency in those areas may be easier or more important to enforce).
That doesn't mean you have to settle for it, though. It's possible to take a progressive approach, where messy metadata comes in, and is then brought into consistency (by humans or machines) with particular standards. Moreover, certain metadata formats (such as MODS) have ways of marking metadata that's informal first-cut versus metadata that's been brought into conformance with a standard, allowing people or programs to go through repositories and catalogs and find and prioritize items to be brought into conformance.
I did something like this with my online books collection, where I started with non-authority-controlled names and no subject terms, and over time brought more names under LC authority controlled, and also over time went from no subjects to automatic subject assignment with somewhat out of date authorities to reviewed subject assignment with current authorities.
The process isn't completely finished (though it's largely done by now), but it doesn't have to be complete to be useful. The metadata includes (implicitly or explicitly) indicators of the quality of name and subject metadata therein, and I can prioritize what to update based on usage and the influx of new and related material.