This incorrect caching behavior can be triggered in editors with
misbehaving undo/redo implementations, and it has been observed in ETA.
The errors that follow from this look very much like several Watsons we
have from Visual Studio, even though we've never been able to prove this
caching failure in Visual Studio. This should let us know if there's
something in Visual Studio violating the ReiteratedVersionNumber
contract, which we would then track down to fix or potentially use as
evidence to remove this secondary cache.