I would also agree that introducing a length filter too early would be detrimental for understanding or interpreting certain experiments. Here are some examples of public datasets that would be affected (even though arguably not too badly):

% of total reads are ribosome protected fragments based on periodicity > Ribotricer phase score?

These stats are taken from the Ribotish results for each read length

Sorry, bit confused here... >30% of reads that are 29nt were considered RPFs? These values were found in riboseq_qc -> ribotish folder?

yes, that folder.