Non-ASCII names in the ChangeLog file in lame 3.99.4 are encoded with different character sets. It doesn't make sense to switch encoding half way through a plain text file.
More precisely, the names
Rogério Brito
Diego "Flameeyes" Pettenò
are encoded using UTF-8, while the names
Sébastien FORESTIER
Ihárosi Wiktor
are encoded using Latin 1. Today I guess using UTF-8 consistently would make the most sense.
The ChangeLog is autogenerated. My assumption is, that it takes whatever is in the version control system. So if the conversion to SVN didn't fix this, it is like it is. We don't modify the history / old entries inside the version control system.