Jon, I don't see where this happens.
Inverse cypher returns the data just as the input of the cypher, however the control is set to HEX-display (very usefull during debuggin), if you drop a control based on that you definitly get something different than you expect, but you shouldn't need to use an additional conversion.
Yes/No, it would be nice if there is some standard of padding... I've looked throught several NIST documents and they only mention padding with 'nulls', not defining an official accepted form of padding. Note that padding depends on the type of block mode you use.
Ton