If images are still in the file they might be referenced by some other object in PDF (e.g. they may be listed in the NameTree, OCG tree, etc.) This is similar to memory leak with managed languages such as C# or JAVA.
You can solve the problem by removing the extra references. A tool such as CosEdit (http://www.pdftron.com/pdfcosedit/downloads.html) may be useful for debugging.
If you send a sample PDF to ‘support at pdftron’, we will look into the file and diagnose the issue for you.