You're right, I tested on Windows.
Removing the db from the equation, the next logical suspect in the jvm, which might need some configuration change, though I can't say exactly what they are at the moment.Parts of the code which are used by encrypt/decrypt is being rewritten, and should be available soon. These changes may solve various issues including those to do with padding.
If you can't wait for these changes, you can try specifying an alternate crypto algorithm, which might be less strict when decrypting - see here for additional details.
For best results, use the Firefox browser..