It seems to me from my understand of the source code, that when the
alter database end backup occurs, the write-locking of the database
header page prevents updates to the rest of the database? (although not
reads?)

This would mean that the fundamental issue that incremental backup is
designed to address, databases getting bigger and bigger and more and
more data changing would be defeated by this mechanism? The bigger and
more active a site gets, the more downtime they will encounter when the
ALTER DATABASE END BACKUP occurs because of merges that need to go back
into the database?

Clearly when the amount of data is down in the low megabytes range, this
isn't a problem, but on busy multi-user sites, this seems to be a big
problem.