Ok. Thanks. I'll test that setting.

The 'emails' setting in nls.php. It's even documented.

Are the any other way to force that all messages are sent using UTF-8 charset?

Our users sent messages that contain texts that would have to be coded 
in different charset. Only way to send these messages is to use UTF-8 
charset.

Why?

The problem is that we have to force the locale to use UTF-8 by 
defining it in horde/config/nls.local.php:
$horde_nls_config['charsets'] ['fi_FI'] = 'UTF-8';

This results that characters are stripped from labels in this piece of code:
return preg_replace('/_([A-Za-z])/', 
$GLOBALS['registry']->nlsconfig->curr_multibyte && 
preg_match('/[\x80-\xff]/', $label) ? '' : '\1', $label);

Which I believe does this: If the translated label contains 8bit 
character and the locale uses multibyte charset eg. UTF-8 then remove 
any character after the underscore.

Stripping blindly the next character after the underscore results 
labels where a character is missing and might even break the resulting 
label if the desired access key is not a ascii character.

No idea what you mean with that. Installing UTF-8 locales? Without 
that, Horde translations won't work at all anyway.
Sorry?
Which problem? This method removes the underscore from labels.
Because there is no reason *not* to use UTF-8. Horde 4 is completely 
using UTF-8 internally and externally and requires UTF-8 support both 
on the server side and client side, which is fortunately not a problem 
anymore in 2012.
What does this have to do access keys?

The actual version is 4.0.15-git.

In order for reproduce this:
1.  Enable UTF-8 to you locale
2.  Localized access key characters are stripped from the string.

The problem or feature lies within the file: 
horde/framework/Core/lib/Horde.php
in function: stripAccessKey

Why we use UTF-8 used instead of the default iso-8859-1? Our users use 
may choose to use english or finnish interface but write letters that 
contain finnish characters (needs iso-8859-1) with euro character 
(needs iso-8859-15) and eg. some cyrillic letters (KOI8-R). Only way 
to include all of these different character sets  in a single message 
is to use UTF-8 as charset.

I don't see this. But version 4.0 is really outdated anyway.

I have translated string:
msgid "Alerts _Log"
msgstr "Hälytys_loki"

Then I have this shown in the program as:
"Hälytysoki(l)"

Note that the letter l is now missing from translated the string: 
"loki" vs "oki(l)"