It *shouldn't* be.  But that doesn't mean it might never leak through. 
  And we can't go munging headers.
You are assuming that sent mail and drafts are static.  They aren't.   
There is no 100% foolproof way of determining whether a message was 
actually sent by IMP or not.

Which leads me back to the previous suggestion: The user has to 
proactively make a choice whether to strip the header out or not.

I think Bcc is an exception because it would never be a real header of 
a real message. We only have them in drafts and sent mail, so these 
messages aren't even "as-is" the actually sent messages. We remove 
that header when sending those messages, so we can as well justify to 
remove it if we forward them.

Forwarded messages can't be munged.  Period.  They HAVE to be sent 
as-is.  That's the entire point of forwarding.

The only viable options are either:
1) Don't save Bcc information in forwarded message. (But now there is 
no complete record of the recipients of the message).
2) Warn user that Bcc information appears in the headers of a 
forwarded message.

when a message (from sentmail folder) is forwarded, any existing Bcc 
header should rather be removed or at least a warning being prompted 
to the user

currently a message being forwarded as an attachment remains untouched 
an thus may contain Bcc lines that are thought to remain secret