Vadim from Oracle says:
Thanks again for providing me with
GEN_ACYCLIC_DFA_INLINE = false
which allowed me to scale my grammar up. Unfortunately, there seems to be a bug -- the lexer swallows some tokens and they never show up in the output stream. Test case:

grammar Test;

id: IDENTIFIER;

IDENTIFIER: 'A' .. 'Z' ( 'A' ..'Z'| '0' .. '9' )*;

WS : (' '|'\r'|'\t'|'\u000C'|'\n') {$channel=HIDDEN;};

When running antlrworks debugger with input text "I" and start rule "id", the parse tree insists on "MismatchedTokenException". The input window in the lower left corner remains empty. Inline DFA case works OK.