I encountered the same problem as LeCroy ( e.g. a second "prompt" interrupt)
when testing the driver: It was sometimes missing one message and losing
synchronization. It was solved by using a counting semaphore for the RX task
(refer to my thesis for details) allowing it to run twice instead of once in
case of two close arrival of messages.
The ARCNET's rate of message passing is so high that it looks impossible to
have a really "real-time" reaction with VxWorks, unless you lock every other
resources ( the tests were done without EPICS...).
regards,
Thierry.