parser generator on top of SAX

Hi,

all Java parser generators I know are using a "pull model" for accessing 
their tokens. For example, they all invoke a method like

     Token nextToken();

While this is fine in general, it is unsuitable, if the token is a SAX 
event, because it leaves either of the following:

     - push the events (aka tokens) into a list and implement nextToken()
       by consuming the list elements; this is bad for large documents
     - Invoke the parser in a separate thread; this is bad for performance
       when parsing a lot of documents

Does anyone of you have recommendations how to fix this problem?


Kind regards,

Jochen