quickbms already uses the latest version of the lzma sdk and uses the most extreme compression setting, so there is not much I can do.
The settings of compression algorithms sometimes don't work as you expect, for example sometimes with some types of files you can get better performances using a normal compression level rather than an extreme one. These settings are hardcoded in quickbms so you can't change them and there is no "brute forcing" mechanism like instead is implemented for the deflate/zlib algorithms (long story).
An easy solution that doesn't involve programming and manual reinjection of the new manually compressed data back in the archive at the exact offset and repeated for all the files to reimport?
If you open quickbms.exe 0.7.5 with a hex editor and go at offset 0xdd25 you will find the byte 0x1b, you can try to replace it with 0x18 and retrying the reimporting but I don't think it will change much because there are many other low level lzma settings used by quickbms.
It's correct because it's NOT a magic, that one is an header with information in it about the parameters of the compression, the dictionary size and so on.