Corpus with readability judgements?

Dear all.

I am looking for a corpus of texts with readability judgements.  The ideal
corpus (which certainly does not exist) would fulfil the following:

- The judgements are from human readers.
- The corpus contains several hundred texts of around 5 - 100 sentences.
- The texts are in German.
- The corpus covers readability aspects on many linguistic levels.
- It contains pairs of parallel texts: easy to read vs. hard to read.

Searching the archives of this mailing list, I only found the REAP project,
which is not exactly what I am looking for.

Greetings
Sven Hartrumpf