The legacy C++ implementation seems to output directly to the C standard output stream stdout, without setting binary mode. On Windows, if the written data includes line-feed 0x0A (\n) bytes, they will get automatically prefixed with a carriage return character 0x0D (\r), resulting in extra bytes in the output.

To reproduce, have some documents in a mongo collection with data containing line feed characters e.g.:

{
    "testfield" : "A\nB"
}

open a command line prompt and run:

mongodump.exe --db mydb --collection mycollection --out testdump
mongodump.exe --db mydb --collection mycollection --out - > testdump.bson

Then observe 1-byte size difference between the files testdump\mydb\mycollection.bson & testdump.bson and confirm that there is an extra 0x0D byte in the latter file.