There are several interesting observations that are made in this paper.
First is the recognition that MAX is not a dataflow language because the object boxes contain a local state violating the constraint of independence of the order of inputs. There is also the problem of dataflow language requirements that variables be modified only once. However, he points out that this is valid because the objects must respond to external requests for action and the order of events is significant.
Second, he notes that there are two types of messages passed around in Max and Pd. One is order dependent and sequential, and the other is duty cycle based. For the order dependent messages, each object's left most input activates the operation with messages passed in the inlets to the right coming before the left. On the duty cycle based messages, all objects update a fixed sized vector in order depending upon interconnectivity.
The other interesting note is that the control and implementation are on two different processors. This is in keeping with my ideas about two display operating systems where one display is used for control and the other is essentially the media rendered in real-time. in this case one display is the computer screen and the other is the output of sound. It is interesting that due to costs, the two processes have merged as computer power has increased.