It would be great to have some interruption points in model loading, to allow for callers to possibly save some time on redundant loads whose result isn't going to be used anyway.
At the very least, a check before validators would be a great start. Presumably, validators themselves could opt into being interruptible as well.