Dr. Frank Hofmann forwarded this announcement in his firstname.lastname@example.org blog indicating the precautions to take in geodatabase handling.:
NIM059333 -Object definitions containing Unicode characters stored in 10.0 personal (access) geodatabases are not persisted correctly.
Description: Non-english Unicode characters used in the definition of items become corrupted in version 10 personal geodatabases when using ArcGIS 10. This error corrupts the source data and is non-recoverable (emphasizing by the author).
Only specific uses of non-english Unicode characters in the following item definitions will cause this behavior to occur (this is not a comprehensive list.): domain names, domain description, domain codes, default values, field aliases, table aliases, subtype names, subtype descriptions, feature dataset names, simple relationship names (but not attributed relationships), replica names, toolboxes names, topology names, geometric network names, network dataset names, terrain names, representation names.
Other uses of non-english Unicode characters, such as in the contents or names of geodatabase tables or feature classes will not be affected by this issue.