This is strange, the behavior on the server side seems to be like that in all versions down to at least 5.1, both MariaDB and MySQL.
It could however happen that the bad value somehow crept in with the upgrade or afterwards.
I'll check upgrade on Wheezy to make sure it doesn't come with the package. Meanwhile, can you check your config files and see where exactly the value is set? Also, are you the only one who has access to the files, can it be that somebody else modified it recently?

– Meanwhile, can you check your config files and see where exactly the value is set?
Will do and get back to you on that.

– Also, are you the only one who has access to the files, can it be that somebody else modified it recently?
Yes, I am the only one with access to this files/machine. I didn't modify them for at least a year ago, so no recent modifications on any config file. (Maybe the upgrade process did something on the config, will investigate that.)

Also refiled at MySQL bug base as http://bugs.mysql.com/bug.php?id=69515

Must have made a mistake (one * 1024 too many) a while ago since I set that value into the config myself. No error from the upgrade progress there but a human error.

Still a bug though, so I'm sort of glad we've run into this.

Thanks again! Set the value to a more acceptable amount and all is working again.

Seems to reproduce also with group_concat_max_len=1024*1024*1024 (2^30), and works with group_concat_max_len=1024*1024*1024-1 (2^30-1).

Perhaps true maximum is 2^30-1 instead of 2^32-1.