One of the factors which is easily overlooked is that virtually all
modern MIDI devices employ some form of data buffering.  This means
that although the data may arrive sequentially, it is stored for
a few milliseconds before being used.  Doing this buys time for the
playing system, enabling it to reorganise the data and align the
events to their precisely correct positions.

As Robbie points out, the delays within the actual playing system
are minute and can be ignored.  This explains why some MIDI devices
appear to cope better than others and it is usually the older ones
which are found lacking.  (As an extreme example of data buffering,
the Disklavier introduces a full half second of delay to enable it
to compensate internally for timing errors of this type.)

Despite this, *when playing just one single piano*, even the worst
implementation of the MIDI standard will prove more than adequate and
will exceed the resolution of a piano roll by a very wide margin.

I'm no lover of MIDI but have to admit that it takes plenty of unfair
blame.  Long ago, I discovered that some poorly designed MIDI devices
choke on simultaneous events.  This was a frustrating situation at the
time.  As the precision of the roll scans increased, the results
sounded worse and worse!

This came about because, unlike real pianists, rolls frequently require
several notes to be struck at precisely the same instant.  I introduced
a small amount of random dither into the MIDI generating software.
This eliminated precisely simultaneous events and solved the problem --
but it wasn't MIDI's fault!

Richard Stibbons
Cromer, England