I don't think you can get around that realistically. Compression works by substituting values with much-shorter-length tokens (just a few bits per token) The most frequent values get the shorter tokens.
The data you are adding likely is using some new values not already present in the table/dictionary, so they would need to be added to the table in addition to your data. You'd have to know where this table is and how it is setup. And then even "inserting" your data would be a very, very sensitive concept, slip on one bit and the entire file beyond the insertion point is completely corrupted.
Even if this were your own compression algorithm and you therefore knew how it worked extensively, you would still find it extremely hard to insert data like this, I think. I do think the only reasonable approach is de-compress, edit, and then re-compress, I'm afraid...