Dan - you'd think so, but we have run into this. Your statement holds if users adhere to best Workspace practices. Where it can (and at times does) fall down is in combination of one or more of the following...
- User(s) keep working in one Workspace for long enough for the generic to get out of date (later version exists on Server)
- Family Tables are unstable and modify/iterate virtually any time an instance is opened
- Family Table is never "locked down" (i.e. set to an ACL non-modifiable State such as "Released")
- User(s) do not use Check Out to proactively "tag" the files that they intend to modify
- User(s) blindly Check Out everything that modifies locally, without understanding why (reasons include but not limited to changes to layer display in a model, circular references, relations in a generic affecting instances, dimensional changes, tolerance changes etc.).
- User(s) Check everything modified in Workspace rather than overwriting unintended modifications (files not Checked Out) with Server side versions
It can, and has happened with let's say a User adding an instance to Rev A Generic and another User adding an instance to Rev B. The Allow Check Out of non latest iterations may not have been set when this occurred, but to satisfy my own curiosity I will probably attempt to add Instances to different (unReleased) Revisions of the generic to see if I can replicate... (I fear that iterations are per Revision)