In my inbox today from ESRI:

"An issue has been found with some uses of non-English Unicode data within the personal geodatabase when upgrading to ArcGIS 10. In some cases, the use of non-English Unicode characters in object names in the geodatabase are corrupted when the personal geodatabase (access file) is upgraded to an ArcGIS 10 geodatabase. This error corrupts the source data and is nonrecoverable. Further details and workarounds are provided in knowledge base article 38149.

A software patch is available for this issue. Download now."