weird comparison between Mallet and Mitao

[lucapareschi]

Results produced by Mallet and by Mitao are very different (see attached file for a comparison). When calculating the topic composition of each document, Mallet considers documents composed by topics in different percentage, and rarely the first topic accounts for more than 50%. Conversely, MITAO often assigns 99% or 100% of a document to a one topic only. Is this an issue with MITAO? Is this a difference between Gensim and Mallet?
comparison_Mallet_Mitao.pdf