On Windows, smb_addmsg() was slow because smb_updatethread() was slow:

So on Windows, calling unlock() on a file region that you don't have locked,
incurs a delay. Likewise, calling lock() on a file region that already have
locked incurs a delay. These delays were seriously throttling message imports
on Windows (e.g. importing echomail or QWK packets). What should take a few
seconds, was taking minutes with noticeable delays. This code has been
performing redundant locks/unlocks for over 15 years, so I can only conclude
that the performance penalty is a new behavior introduced in recent version(s)
of Windows.