Importing a MIDI into a notation program usually works best when the file itself has been output from a sequencer program, and the note lengths are accurate.

Where the file is created by actually, playing the tune on a MIDI instrument, the notes are usually shorter than expected, because of the need to leave slight gaps between them. This is what gives rise to all those 32nd rests. Notes are also of uneven length and variable start time compared to a metronome.

These are the files which cause most problems in import, but are the ones which sound most realistic.

The problems can be reduced by quantisation, as noted above, but this is not a complete remedy. I have been experimenting with an improved 'smart' quantisation, but haven't really solved the problem yet.