Two central design principles of the BIT are:
1. Fail fast
2. Reruns are cheap (they pick up where they left off)
Relevant to your question is also the fact that the tool uploads files in transactional batches (for performance reasons) these batches a variably sized and can be as small as a single file, or as large as 100 files (by default - this can be overridden via configuration).
Putting this all together in the context of your question, what this means is that when a file unexpectedly fails to upload, the tool will terminate the import, and everything that had previously been written to the repository will remain in the repository, *except* for any files that happened to be in the same batch as the file that failed (these files will be rolled back). The reason for the failure will be written to the alfresco.log.
Once the issue has been remedied, the correct action is to simply rerun the exact same import, with “replace existing files” unchecked (disabled / turned off). The tool will quickly skip over the files it already imported, and pick up where it left off.